2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/icmp.h>
45 #include <linux/icmpv6.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_dissector.h>
83 #include <net/bonding.h>
84 #include <net/bond_3ad.h>
85 #include <net/bond_alb.h>
86 #if IS_ENABLED(CONFIG_TLS_DEVICE)
90 #include "bonding_priv.h"
92 /*---------------------------- Module parameters ----------------------------*/
94 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
96 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
97 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
98 static int num_peer_notif = 1;
101 static int downdelay;
102 static int use_carrier = 1;
104 static char *primary;
105 static char *primary_reselect;
106 static char *lacp_rate;
107 static int min_links;
108 static char *ad_select;
109 static char *xmit_hash_policy;
110 static int arp_interval;
111 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
112 static char *arp_validate;
113 static char *arp_all_targets;
114 static char *fail_over_mac;
115 static int all_slaves_active;
116 static struct bond_params bonding_defaults;
117 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
118 static int packets_per_slave = 1;
119 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
121 module_param(max_bonds, int, 0);
122 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
123 module_param(tx_queues, int, 0);
124 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
125 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
126 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
127 "failover event (alias of num_unsol_na)");
128 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
129 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
130 "failover event (alias of num_grat_arp)");
131 module_param(miimon, int, 0);
132 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
133 module_param(updelay, int, 0);
134 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
135 module_param(downdelay, int, 0);
136 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
138 module_param(use_carrier, int, 0);
139 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
140 "0 for off, 1 for on (default)");
141 module_param(mode, charp, 0);
142 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
143 "1 for active-backup, 2 for balance-xor, "
144 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
145 "6 for balance-alb");
146 module_param(primary, charp, 0);
147 MODULE_PARM_DESC(primary, "Primary network device to use");
148 module_param(primary_reselect, charp, 0);
149 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
151 "0 for always (default), "
152 "1 for only if speed of primary is "
154 "2 for only on active slave "
156 module_param(lacp_rate, charp, 0);
157 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
158 "0 for slow, 1 for fast");
159 module_param(ad_select, charp, 0);
160 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
161 "0 for stable (default), 1 for bandwidth, "
163 module_param(min_links, int, 0);
164 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
166 module_param(xmit_hash_policy, charp, 0);
167 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
168 "0 for layer 2 (default), 1 for layer 3+4, "
169 "2 for layer 2+3, 3 for encap layer 2+3, "
170 "4 for encap layer 3+4, 5 for vlan+srcmac");
171 module_param(arp_interval, int, 0);
172 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
173 module_param_array(arp_ip_target, charp, NULL, 0);
174 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
175 module_param(arp_validate, charp, 0);
176 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
177 "0 for none (default), 1 for active, "
178 "2 for backup, 3 for all");
179 module_param(arp_all_targets, charp, 0);
180 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
181 module_param(fail_over_mac, charp, 0);
182 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
183 "the same MAC; 0 for none (default), "
184 "1 for active, 2 for follow");
185 module_param(all_slaves_active, int, 0);
186 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
187 "by setting active flag for all slaves; "
188 "0 for never (default), 1 for always.");
189 module_param(resend_igmp, int, 0);
190 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
192 module_param(packets_per_slave, int, 0);
193 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
194 "mode; 0 for a random slave, 1 packet per "
195 "slave (default), >1 packets per slave.");
196 module_param(lp_interval, uint, 0);
197 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
198 "the bonding driver sends learning packets to "
199 "each slaves peer switch. The default is 1.");
201 /*----------------------------- Global variables ----------------------------*/
203 #ifdef CONFIG_NET_POLL_CONTROLLER
204 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
207 unsigned int bond_net_id __read_mostly;
209 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
211 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
212 .offset = offsetof(struct flow_keys, control),
215 .key_id = FLOW_DISSECTOR_KEY_BASIC,
216 .offset = offsetof(struct flow_keys, basic),
219 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
220 .offset = offsetof(struct flow_keys, addrs.v4addrs),
223 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
224 .offset = offsetof(struct flow_keys, addrs.v6addrs),
227 .key_id = FLOW_DISSECTOR_KEY_TIPC,
228 .offset = offsetof(struct flow_keys, addrs.tipckey),
231 .key_id = FLOW_DISSECTOR_KEY_PORTS,
232 .offset = offsetof(struct flow_keys, ports),
235 .key_id = FLOW_DISSECTOR_KEY_ICMP,
236 .offset = offsetof(struct flow_keys, icmp),
239 .key_id = FLOW_DISSECTOR_KEY_VLAN,
240 .offset = offsetof(struct flow_keys, vlan),
243 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
244 .offset = offsetof(struct flow_keys, tags),
247 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
248 .offset = offsetof(struct flow_keys, keyid),
252 static struct flow_dissector flow_keys_bonding __read_mostly;
254 /*-------------------------- Forward declarations ---------------------------*/
256 static int bond_init(struct net_device *bond_dev);
257 static void bond_uninit(struct net_device *bond_dev);
258 static void bond_get_stats(struct net_device *bond_dev,
259 struct rtnl_link_stats64 *stats);
260 static void bond_slave_arr_handler(struct work_struct *work);
261 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
263 static void bond_netdev_notify_work(struct work_struct *work);
265 /*---------------------------- General routines -----------------------------*/
267 const char *bond_mode_name(int mode)
269 static const char *names[] = {
270 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
271 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
272 [BOND_MODE_XOR] = "load balancing (xor)",
273 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
274 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
275 [BOND_MODE_TLB] = "transmit load balancing",
276 [BOND_MODE_ALB] = "adaptive load balancing",
279 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
295 skb->dev = slave_dev;
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
304 return dev_queue_xmit(skb);
307 bool bond_sk_check(struct bonding *bond)
309 switch (BOND_MODE(bond)) {
310 case BOND_MODE_8023AD:
312 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
320 /*---------------------------------- VLAN -----------------------------------*/
322 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
323 * We don't protect the slave list iteration with a lock because:
324 * a. This operation is performed in IOCTL context,
325 * b. The operation is protected by the RTNL semaphore in the 8021q code,
326 * c. Holding a lock with BH disabled while directly calling a base driver
327 * entry point is generally a BAD idea.
329 * The design of synchronization/protection for this operation in the 8021q
330 * module is good for one or more VLAN devices over a single physical device
331 * and cannot be extended for a teaming solution like bonding, so there is a
332 * potential race condition here where a net device from the vlan group might
333 * be referenced (either by a base driver or the 8021q code) while it is being
334 * removed from the system. However, it turns out we're not making matters
335 * worse, and if it works for regular VLAN usage it will work here too.
339 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
340 * @bond_dev: bonding net device that got called
341 * @proto: network protocol ID
342 * @vid: vlan id being added
344 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
345 __be16 proto, u16 vid)
347 struct bonding *bond = netdev_priv(bond_dev);
348 struct slave *slave, *rollback_slave;
349 struct list_head *iter;
352 bond_for_each_slave(bond, slave, iter) {
353 res = vlan_vid_add(slave->dev, proto, vid);
361 /* unwind to the slave that failed */
362 bond_for_each_slave(bond, rollback_slave, iter) {
363 if (rollback_slave == slave)
366 vlan_vid_del(rollback_slave->dev, proto, vid);
373 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
374 * @bond_dev: bonding net device that got called
375 * @proto: network protocol ID
376 * @vid: vlan id being removed
378 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
379 __be16 proto, u16 vid)
381 struct bonding *bond = netdev_priv(bond_dev);
382 struct list_head *iter;
385 bond_for_each_slave(bond, slave, iter)
386 vlan_vid_del(slave->dev, proto, vid);
388 if (bond_is_lb(bond))
389 bond_alb_clear_vlan(bond, vid);
394 /*---------------------------------- XFRM -----------------------------------*/
396 #ifdef CONFIG_XFRM_OFFLOAD
398 * bond_ipsec_add_sa - program device with a security association
399 * @xs: pointer to transformer state struct
401 static int bond_ipsec_add_sa(struct xfrm_state *xs)
403 struct net_device *bond_dev = xs->xso.dev;
404 struct bonding *bond;
410 bond = netdev_priv(bond_dev);
411 slave = rcu_dereference(bond->curr_active_slave);
412 xs->xso.real_dev = slave->dev;
415 if (!(slave->dev->xfrmdev_ops
416 && slave->dev->xfrmdev_ops->xdo_dev_state_add)) {
417 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
421 return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
425 * bond_ipsec_del_sa - clear out this specific SA
426 * @xs: pointer to transformer state struct
428 static void bond_ipsec_del_sa(struct xfrm_state *xs)
430 struct net_device *bond_dev = xs->xso.dev;
431 struct bonding *bond;
437 bond = netdev_priv(bond_dev);
438 slave = rcu_dereference(bond->curr_active_slave);
443 xs->xso.real_dev = slave->dev;
445 if (!(slave->dev->xfrmdev_ops
446 && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) {
447 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
451 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
455 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
456 * @skb: current data packet
457 * @xs: pointer to transformer state struct
459 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
461 struct net_device *bond_dev = xs->xso.dev;
462 struct bonding *bond = netdev_priv(bond_dev);
463 struct slave *curr_active = rcu_dereference(bond->curr_active_slave);
464 struct net_device *slave_dev = curr_active->dev;
466 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
469 if (!(slave_dev->xfrmdev_ops
470 && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) {
471 slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__);
475 xs->xso.real_dev = slave_dev;
476 return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
479 static const struct xfrmdev_ops bond_xfrmdev_ops = {
480 .xdo_dev_state_add = bond_ipsec_add_sa,
481 .xdo_dev_state_delete = bond_ipsec_del_sa,
482 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
484 #endif /* CONFIG_XFRM_OFFLOAD */
486 /*------------------------------- Link status -------------------------------*/
488 /* Set the carrier state for the master according to the state of its
489 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
490 * do special 802.3ad magic.
492 * Returns zero if carrier state does not change, nonzero if it does.
494 int bond_set_carrier(struct bonding *bond)
496 struct list_head *iter;
499 if (!bond_has_slaves(bond))
502 if (BOND_MODE(bond) == BOND_MODE_8023AD)
503 return bond_3ad_set_carrier(bond);
505 bond_for_each_slave(bond, slave, iter) {
506 if (slave->link == BOND_LINK_UP) {
507 if (!netif_carrier_ok(bond->dev)) {
508 netif_carrier_on(bond->dev);
516 if (netif_carrier_ok(bond->dev)) {
517 netif_carrier_off(bond->dev);
523 /* Get link speed and duplex from the slave's base driver
524 * using ethtool. If for some reason the call fails or the
525 * values are invalid, set speed and duplex to -1,
526 * and return. Return 1 if speed or duplex settings are
527 * UNKNOWN; 0 otherwise.
529 static int bond_update_speed_duplex(struct slave *slave)
531 struct net_device *slave_dev = slave->dev;
532 struct ethtool_link_ksettings ecmd;
535 slave->speed = SPEED_UNKNOWN;
536 slave->duplex = DUPLEX_UNKNOWN;
538 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
541 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
543 switch (ecmd.base.duplex) {
551 slave->speed = ecmd.base.speed;
552 slave->duplex = ecmd.base.duplex;
557 const char *bond_slave_link_status(s8 link)
573 /* if <dev> supports MII link status reporting, check its link status.
575 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
576 * depending upon the setting of the use_carrier parameter.
578 * Return either BMSR_LSTATUS, meaning that the link is up (or we
579 * can't tell and just pretend it is), or 0, meaning that the link is
582 * If reporting is non-zero, instead of faking link up, return -1 if
583 * both ETHTOOL and MII ioctls fail (meaning the device does not
584 * support them). If use_carrier is set, return whatever it says.
585 * It'd be nice if there was a good way to tell if a driver supports
586 * netif_carrier, but there really isn't.
588 static int bond_check_dev_link(struct bonding *bond,
589 struct net_device *slave_dev, int reporting)
591 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
592 int (*ioctl)(struct net_device *, struct ifreq *, int);
594 struct mii_ioctl_data *mii;
596 if (!reporting && !netif_running(slave_dev))
599 if (bond->params.use_carrier)
600 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
602 /* Try to get link status using Ethtool first. */
603 if (slave_dev->ethtool_ops->get_link)
604 return slave_dev->ethtool_ops->get_link(slave_dev) ?
607 /* Ethtool can't be used, fallback to MII ioctls. */
608 ioctl = slave_ops->ndo_do_ioctl;
610 /* TODO: set pointer to correct ioctl on a per team member
611 * bases to make this more efficient. that is, once
612 * we determine the correct ioctl, we will always
613 * call it and not the others for that team
617 /* We cannot assume that SIOCGMIIPHY will also read a
618 * register; not all network drivers (e.g., e100)
622 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
623 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
625 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
626 mii->reg_num = MII_BMSR;
627 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
628 return mii->val_out & BMSR_LSTATUS;
632 /* If reporting, report that either there's no dev->do_ioctl,
633 * or both SIOCGMIIREG and get_link failed (meaning that we
634 * cannot report link status). If not reporting, pretend
637 return reporting ? -1 : BMSR_LSTATUS;
640 /*----------------------------- Multicast list ------------------------------*/
642 /* Push the promiscuity flag down to appropriate slaves */
643 static int bond_set_promiscuity(struct bonding *bond, int inc)
645 struct list_head *iter;
648 if (bond_uses_primary(bond)) {
649 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
652 err = dev_set_promiscuity(curr_active->dev, inc);
656 bond_for_each_slave(bond, slave, iter) {
657 err = dev_set_promiscuity(slave->dev, inc);
665 /* Push the allmulti flag down to all slaves */
666 static int bond_set_allmulti(struct bonding *bond, int inc)
668 struct list_head *iter;
671 if (bond_uses_primary(bond)) {
672 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
675 err = dev_set_allmulti(curr_active->dev, inc);
679 bond_for_each_slave(bond, slave, iter) {
680 err = dev_set_allmulti(slave->dev, inc);
688 /* Retrieve the list of registered multicast addresses for the bonding
689 * device and retransmit an IGMP JOIN request to the current active
692 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
694 struct bonding *bond = container_of(work, struct bonding,
697 if (!rtnl_trylock()) {
698 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
701 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
703 if (bond->igmp_retrans > 1) {
704 bond->igmp_retrans--;
705 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
710 /* Flush bond's hardware addresses from slave */
711 static void bond_hw_addr_flush(struct net_device *bond_dev,
712 struct net_device *slave_dev)
714 struct bonding *bond = netdev_priv(bond_dev);
716 dev_uc_unsync(slave_dev, bond_dev);
717 dev_mc_unsync(slave_dev, bond_dev);
719 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
720 /* del lacpdu mc addr from mc list */
721 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
723 dev_mc_del(slave_dev, lacpdu_multicast);
727 /*--------------------------- Active slave change ---------------------------*/
729 /* Update the hardware address list and promisc/allmulti for the new and
730 * old active slaves (if any). Modes that are not using primary keep all
731 * slaves up date at all times; only the modes that use primary need to call
732 * this function to swap these settings during a failover.
734 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
735 struct slave *old_active)
738 if (bond->dev->flags & IFF_PROMISC)
739 dev_set_promiscuity(old_active->dev, -1);
741 if (bond->dev->flags & IFF_ALLMULTI)
742 dev_set_allmulti(old_active->dev, -1);
744 bond_hw_addr_flush(bond->dev, old_active->dev);
748 /* FIXME: Signal errors upstream. */
749 if (bond->dev->flags & IFF_PROMISC)
750 dev_set_promiscuity(new_active->dev, 1);
752 if (bond->dev->flags & IFF_ALLMULTI)
753 dev_set_allmulti(new_active->dev, 1);
755 netif_addr_lock_bh(bond->dev);
756 dev_uc_sync(new_active->dev, bond->dev);
757 dev_mc_sync(new_active->dev, bond->dev);
758 netif_addr_unlock_bh(bond->dev);
763 * bond_set_dev_addr - clone slave's address to bond
764 * @bond_dev: bond net device
765 * @slave_dev: slave net device
767 * Should be called with RTNL held.
769 static int bond_set_dev_addr(struct net_device *bond_dev,
770 struct net_device *slave_dev)
774 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
775 bond_dev, slave_dev, slave_dev->addr_len);
776 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
780 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
781 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
782 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
786 static struct slave *bond_get_old_active(struct bonding *bond,
787 struct slave *new_active)
790 struct list_head *iter;
792 bond_for_each_slave(bond, slave, iter) {
793 if (slave == new_active)
796 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
803 /* bond_do_fail_over_mac
805 * Perform special MAC address swapping for fail_over_mac settings
809 static void bond_do_fail_over_mac(struct bonding *bond,
810 struct slave *new_active,
811 struct slave *old_active)
813 u8 tmp_mac[MAX_ADDR_LEN];
814 struct sockaddr_storage ss;
817 switch (bond->params.fail_over_mac) {
818 case BOND_FOM_ACTIVE:
820 rv = bond_set_dev_addr(bond->dev, new_active->dev);
822 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
826 case BOND_FOM_FOLLOW:
827 /* if new_active && old_active, swap them
828 * if just old_active, do nothing (going to no active slave)
829 * if just new_active, set new_active to bond's MAC
835 old_active = bond_get_old_active(bond, new_active);
838 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
839 new_active->dev->addr_len);
840 bond_hw_addr_copy(ss.__data,
841 old_active->dev->dev_addr,
842 old_active->dev->addr_len);
843 ss.ss_family = new_active->dev->type;
845 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
846 bond->dev->addr_len);
847 ss.ss_family = bond->dev->type;
850 rv = dev_set_mac_address(new_active->dev,
851 (struct sockaddr *)&ss, NULL);
853 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
861 bond_hw_addr_copy(ss.__data, tmp_mac,
862 new_active->dev->addr_len);
863 ss.ss_family = old_active->dev->type;
865 rv = dev_set_mac_address(old_active->dev,
866 (struct sockaddr *)&ss, NULL);
868 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
873 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
874 bond->params.fail_over_mac);
880 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
882 struct slave *prim = rtnl_dereference(bond->primary_slave);
883 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
885 if (!prim || prim->link != BOND_LINK_UP) {
886 if (!curr || curr->link != BOND_LINK_UP)
891 if (bond->force_primary) {
892 bond->force_primary = false;
896 if (!curr || curr->link != BOND_LINK_UP)
899 /* At this point, prim and curr are both up */
900 switch (bond->params.primary_reselect) {
901 case BOND_PRI_RESELECT_ALWAYS:
903 case BOND_PRI_RESELECT_BETTER:
904 if (prim->speed < curr->speed)
906 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
909 case BOND_PRI_RESELECT_FAILURE:
912 netdev_err(bond->dev, "impossible primary_reselect %d\n",
913 bond->params.primary_reselect);
919 * bond_find_best_slave - select the best available slave to be the active one
920 * @bond: our bonding struct
922 static struct slave *bond_find_best_slave(struct bonding *bond)
924 struct slave *slave, *bestslave = NULL;
925 struct list_head *iter;
926 int mintime = bond->params.updelay;
928 slave = bond_choose_primary_or_current(bond);
932 bond_for_each_slave(bond, slave, iter) {
933 if (slave->link == BOND_LINK_UP)
935 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
936 slave->delay < mintime) {
937 mintime = slave->delay;
945 static bool bond_should_notify_peers(struct bonding *bond)
950 slave = rcu_dereference(bond->curr_active_slave);
953 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
954 slave ? slave->dev->name : "NULL");
956 if (!slave || !bond->send_peer_notif ||
957 bond->send_peer_notif %
958 max(1, bond->params.peer_notif_delay) != 0 ||
959 !netif_carrier_ok(bond->dev) ||
960 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
967 * bond_change_active_slave - change the active slave into the specified one
968 * @bond: our bonding struct
969 * @new_active: the new slave to make the active one
971 * Set the new slave to the bond's settings and unset them on the old
973 * Setting include flags, mc-list, promiscuity, allmulti, etc.
975 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
976 * because it is apparently the best available slave we have, even though its
977 * updelay hasn't timed out yet.
979 * Caller must hold RTNL.
981 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
983 struct slave *old_active;
987 old_active = rtnl_dereference(bond->curr_active_slave);
989 if (old_active == new_active)
992 #ifdef CONFIG_XFRM_OFFLOAD
993 if (old_active && bond->xs)
994 bond_ipsec_del_sa(bond->xs);
995 #endif /* CONFIG_XFRM_OFFLOAD */
998 new_active->last_link_up = jiffies;
1000 if (new_active->link == BOND_LINK_BACK) {
1001 if (bond_uses_primary(bond)) {
1002 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1003 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1006 new_active->delay = 0;
1007 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1008 BOND_SLAVE_NOTIFY_NOW);
1010 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1011 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1013 if (bond_is_lb(bond))
1014 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1016 if (bond_uses_primary(bond)) {
1017 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1022 if (bond_uses_primary(bond))
1023 bond_hw_addr_swap(bond, new_active, old_active);
1025 if (bond_is_lb(bond)) {
1026 bond_alb_handle_active_change(bond, new_active);
1028 bond_set_slave_inactive_flags(old_active,
1029 BOND_SLAVE_NOTIFY_NOW);
1031 bond_set_slave_active_flags(new_active,
1032 BOND_SLAVE_NOTIFY_NOW);
1034 rcu_assign_pointer(bond->curr_active_slave, new_active);
1037 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1039 bond_set_slave_inactive_flags(old_active,
1040 BOND_SLAVE_NOTIFY_NOW);
1043 bool should_notify_peers = false;
1045 bond_set_slave_active_flags(new_active,
1046 BOND_SLAVE_NOTIFY_NOW);
1048 if (bond->params.fail_over_mac)
1049 bond_do_fail_over_mac(bond, new_active,
1052 if (netif_running(bond->dev)) {
1053 bond->send_peer_notif =
1054 bond->params.num_peer_notif *
1055 max(1, bond->params.peer_notif_delay);
1056 should_notify_peers =
1057 bond_should_notify_peers(bond);
1060 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1061 if (should_notify_peers) {
1062 bond->send_peer_notif--;
1063 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1069 #ifdef CONFIG_XFRM_OFFLOAD
1070 if (new_active && bond->xs) {
1071 xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
1072 bond_ipsec_add_sa(bond->xs);
1074 #endif /* CONFIG_XFRM_OFFLOAD */
1076 /* resend IGMP joins since active slave has changed or
1077 * all were sent on curr_active_slave.
1078 * resend only if bond is brought up with the affected
1079 * bonding modes and the retransmission is enabled
1081 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1082 ((bond_uses_primary(bond) && new_active) ||
1083 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1084 bond->igmp_retrans = bond->params.resend_igmp;
1085 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1090 * bond_select_active_slave - select a new active slave, if needed
1091 * @bond: our bonding struct
1093 * This functions should be called when one of the following occurs:
1094 * - The old curr_active_slave has been released or lost its link.
1095 * - The primary_slave has got its link back.
1096 * - A slave has got its link back and there's no old curr_active_slave.
1098 * Caller must hold RTNL.
1100 void bond_select_active_slave(struct bonding *bond)
1102 struct slave *best_slave;
1107 best_slave = bond_find_best_slave(bond);
1108 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1109 bond_change_active_slave(bond, best_slave);
1110 rv = bond_set_carrier(bond);
1114 if (netif_carrier_ok(bond->dev))
1115 netdev_info(bond->dev, "active interface up!\n");
1117 netdev_info(bond->dev, "now running without any active interface!\n");
1121 #ifdef CONFIG_NET_POLL_CONTROLLER
1122 static inline int slave_enable_netpoll(struct slave *slave)
1127 np = kzalloc(sizeof(*np), GFP_KERNEL);
1132 err = __netpoll_setup(np, slave->dev);
1141 static inline void slave_disable_netpoll(struct slave *slave)
1143 struct netpoll *np = slave->np;
1153 static void bond_poll_controller(struct net_device *bond_dev)
1155 struct bonding *bond = netdev_priv(bond_dev);
1156 struct slave *slave = NULL;
1157 struct list_head *iter;
1158 struct ad_info ad_info;
1160 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1161 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1164 bond_for_each_slave_rcu(bond, slave, iter) {
1165 if (!bond_slave_is_up(slave))
1168 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1169 struct aggregator *agg =
1170 SLAVE_AD_INFO(slave)->port.aggregator;
1173 agg->aggregator_identifier != ad_info.aggregator_id)
1177 netpoll_poll_dev(slave->dev);
1181 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1183 struct bonding *bond = netdev_priv(bond_dev);
1184 struct list_head *iter;
1185 struct slave *slave;
1187 bond_for_each_slave(bond, slave, iter)
1188 if (bond_slave_is_up(slave))
1189 slave_disable_netpoll(slave);
1192 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1194 struct bonding *bond = netdev_priv(dev);
1195 struct list_head *iter;
1196 struct slave *slave;
1199 bond_for_each_slave(bond, slave, iter) {
1200 err = slave_enable_netpoll(slave);
1202 bond_netpoll_cleanup(dev);
1209 static inline int slave_enable_netpoll(struct slave *slave)
1213 static inline void slave_disable_netpoll(struct slave *slave)
1216 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1221 /*---------------------------------- IOCTL ----------------------------------*/
1223 static netdev_features_t bond_fix_features(struct net_device *dev,
1224 netdev_features_t features)
1226 struct bonding *bond = netdev_priv(dev);
1227 struct list_head *iter;
1228 netdev_features_t mask;
1229 struct slave *slave;
1231 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1232 if (bond_sk_check(bond))
1233 features |= BOND_TLS_FEATURES;
1235 features &= ~BOND_TLS_FEATURES;
1240 features &= ~NETIF_F_ONE_FOR_ALL;
1241 features |= NETIF_F_ALL_FOR_ALL;
1243 bond_for_each_slave(bond, slave, iter) {
1244 features = netdev_increment_features(features,
1245 slave->dev->features,
1248 features = netdev_add_tso_features(features, mask);
1253 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1254 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1255 NETIF_F_HIGHDMA | NETIF_F_LRO)
1257 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1258 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1260 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1261 NETIF_F_GSO_SOFTWARE)
1264 static void bond_compute_features(struct bonding *bond)
1266 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1267 IFF_XMIT_DST_RELEASE_PERM;
1268 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1269 netdev_features_t enc_features = BOND_ENC_FEATURES;
1270 #ifdef CONFIG_XFRM_OFFLOAD
1271 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1272 #endif /* CONFIG_XFRM_OFFLOAD */
1273 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1274 struct net_device *bond_dev = bond->dev;
1275 struct list_head *iter;
1276 struct slave *slave;
1277 unsigned short max_hard_header_len = ETH_HLEN;
1278 unsigned int gso_max_size = GSO_MAX_SIZE;
1279 u16 gso_max_segs = GSO_MAX_SEGS;
1281 if (!bond_has_slaves(bond))
1283 vlan_features &= NETIF_F_ALL_FOR_ALL;
1284 mpls_features &= NETIF_F_ALL_FOR_ALL;
1286 bond_for_each_slave(bond, slave, iter) {
1287 vlan_features = netdev_increment_features(vlan_features,
1288 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1290 enc_features = netdev_increment_features(enc_features,
1291 slave->dev->hw_enc_features,
1294 #ifdef CONFIG_XFRM_OFFLOAD
1295 xfrm_features = netdev_increment_features(xfrm_features,
1296 slave->dev->hw_enc_features,
1297 BOND_XFRM_FEATURES);
1298 #endif /* CONFIG_XFRM_OFFLOAD */
1300 mpls_features = netdev_increment_features(mpls_features,
1301 slave->dev->mpls_features,
1302 BOND_MPLS_FEATURES);
1304 dst_release_flag &= slave->dev->priv_flags;
1305 if (slave->dev->hard_header_len > max_hard_header_len)
1306 max_hard_header_len = slave->dev->hard_header_len;
1308 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1309 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1311 bond_dev->hard_header_len = max_hard_header_len;
1314 bond_dev->vlan_features = vlan_features;
1315 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1316 NETIF_F_HW_VLAN_CTAG_TX |
1317 NETIF_F_HW_VLAN_STAG_TX;
1318 #ifdef CONFIG_XFRM_OFFLOAD
1319 bond_dev->hw_enc_features |= xfrm_features;
1320 #endif /* CONFIG_XFRM_OFFLOAD */
1321 bond_dev->mpls_features = mpls_features;
1322 bond_dev->gso_max_segs = gso_max_segs;
1323 netif_set_gso_max_size(bond_dev, gso_max_size);
1325 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1326 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1327 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1328 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1330 netdev_change_features(bond_dev);
1333 static void bond_setup_by_slave(struct net_device *bond_dev,
1334 struct net_device *slave_dev)
1336 bond_dev->header_ops = slave_dev->header_ops;
1338 bond_dev->type = slave_dev->type;
1339 bond_dev->hard_header_len = slave_dev->hard_header_len;
1340 bond_dev->needed_headroom = slave_dev->needed_headroom;
1341 bond_dev->addr_len = slave_dev->addr_len;
1343 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1344 slave_dev->addr_len);
1347 /* On bonding slaves other than the currently active slave, suppress
1348 * duplicates except for alb non-mcast/bcast.
1350 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1351 struct slave *slave,
1352 struct bonding *bond)
1354 if (bond_is_slave_inactive(slave)) {
1355 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1356 skb->pkt_type != PACKET_BROADCAST &&
1357 skb->pkt_type != PACKET_MULTICAST)
1364 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1366 struct sk_buff *skb = *pskb;
1367 struct slave *slave;
1368 struct bonding *bond;
1369 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1371 int ret = RX_HANDLER_ANOTHER;
1373 skb = skb_share_check(skb, GFP_ATOMIC);
1375 return RX_HANDLER_CONSUMED;
1379 slave = bond_slave_get_rcu(skb->dev);
1382 recv_probe = READ_ONCE(bond->recv_probe);
1384 ret = recv_probe(skb, bond, slave);
1385 if (ret == RX_HANDLER_CONSUMED) {
1392 * For packets determined by bond_should_deliver_exact_match() call to
1393 * be suppressed we want to make an exception for link-local packets.
1394 * This is necessary for e.g. LLDP daemons to be able to monitor
1395 * inactive slave links without being forced to bind to them
1398 * At the same time, packets that are passed to the bonding master
1399 * (including link-local ones) can have their originating interface
1400 * determined via PACKET_ORIGDEV socket option.
1402 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1403 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1404 return RX_HANDLER_PASS;
1405 return RX_HANDLER_EXACT;
1408 skb->dev = bond->dev;
1410 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1411 netif_is_bridge_port(bond->dev) &&
1412 skb->pkt_type == PACKET_HOST) {
1414 if (unlikely(skb_cow_head(skb,
1415 skb->data - skb_mac_header(skb)))) {
1417 return RX_HANDLER_CONSUMED;
1419 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1420 bond->dev->addr_len);
1426 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1428 switch (BOND_MODE(bond)) {
1429 case BOND_MODE_ROUNDROBIN:
1430 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1431 case BOND_MODE_ACTIVEBACKUP:
1432 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1433 case BOND_MODE_BROADCAST:
1434 return NETDEV_LAG_TX_TYPE_BROADCAST;
1436 case BOND_MODE_8023AD:
1437 return NETDEV_LAG_TX_TYPE_HASH;
1439 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1443 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1444 enum netdev_lag_tx_type type)
1446 if (type != NETDEV_LAG_TX_TYPE_HASH)
1447 return NETDEV_LAG_HASH_NONE;
1449 switch (bond->params.xmit_policy) {
1450 case BOND_XMIT_POLICY_LAYER2:
1451 return NETDEV_LAG_HASH_L2;
1452 case BOND_XMIT_POLICY_LAYER34:
1453 return NETDEV_LAG_HASH_L34;
1454 case BOND_XMIT_POLICY_LAYER23:
1455 return NETDEV_LAG_HASH_L23;
1456 case BOND_XMIT_POLICY_ENCAP23:
1457 return NETDEV_LAG_HASH_E23;
1458 case BOND_XMIT_POLICY_ENCAP34:
1459 return NETDEV_LAG_HASH_E34;
1460 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1461 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1463 return NETDEV_LAG_HASH_UNKNOWN;
1467 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1468 struct netlink_ext_ack *extack)
1470 struct netdev_lag_upper_info lag_upper_info;
1471 enum netdev_lag_tx_type type;
1473 type = bond_lag_tx_type(bond);
1474 lag_upper_info.tx_type = type;
1475 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1477 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1478 &lag_upper_info, extack);
1481 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1483 netdev_upper_dev_unlink(slave->dev, bond->dev);
1484 slave->dev->flags &= ~IFF_SLAVE;
1487 static void slave_kobj_release(struct kobject *kobj)
1489 struct slave *slave = to_slave(kobj);
1490 struct bonding *bond = bond_get_bond_by_slave(slave);
1492 cancel_delayed_work_sync(&slave->notify_work);
1493 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1494 kfree(SLAVE_AD_INFO(slave));
1499 static struct kobj_type slave_ktype = {
1500 .release = slave_kobj_release,
1502 .sysfs_ops = &slave_sysfs_ops,
1506 static int bond_kobj_init(struct slave *slave)
1510 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1511 &(slave->dev->dev.kobj), "bonding_slave");
1513 kobject_put(&slave->kobj);
1518 static struct slave *bond_alloc_slave(struct bonding *bond,
1519 struct net_device *slave_dev)
1521 struct slave *slave = NULL;
1523 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1528 slave->dev = slave_dev;
1530 if (bond_kobj_init(slave))
1533 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1534 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1536 if (!SLAVE_AD_INFO(slave)) {
1537 kobject_put(&slave->kobj);
1541 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1546 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1548 info->bond_mode = BOND_MODE(bond);
1549 info->miimon = bond->params.miimon;
1550 info->num_slaves = bond->slave_cnt;
1553 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1555 strcpy(info->slave_name, slave->dev->name);
1556 info->link = slave->link;
1557 info->state = bond_slave_state(slave);
1558 info->link_failure_count = slave->link_failure_count;
1561 static void bond_netdev_notify_work(struct work_struct *_work)
1563 struct slave *slave = container_of(_work, struct slave,
1566 if (rtnl_trylock()) {
1567 struct netdev_bonding_info binfo;
1569 bond_fill_ifslave(slave, &binfo.slave);
1570 bond_fill_ifbond(slave->bond, &binfo.master);
1571 netdev_bonding_info_change(slave->dev, &binfo);
1574 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1578 void bond_queue_slave_event(struct slave *slave)
1580 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1583 void bond_lower_state_changed(struct slave *slave)
1585 struct netdev_lag_lower_state_info info;
1587 info.link_up = slave->link == BOND_LINK_UP ||
1588 slave->link == BOND_LINK_FAIL;
1589 info.tx_enabled = bond_is_active_slave(slave);
1590 netdev_lower_state_changed(slave->dev, &info);
1593 /* enslave device <slave> to bond device <master> */
1594 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1595 struct netlink_ext_ack *extack)
1597 struct bonding *bond = netdev_priv(bond_dev);
1598 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1599 struct slave *new_slave = NULL, *prev_slave;
1600 struct sockaddr_storage ss;
1604 if (!bond->params.use_carrier &&
1605 slave_dev->ethtool_ops->get_link == NULL &&
1606 slave_ops->ndo_do_ioctl == NULL) {
1607 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1610 /* already in-use? */
1611 if (netdev_is_rx_handler_busy(slave_dev)) {
1612 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1613 slave_err(bond_dev, slave_dev,
1614 "Error: Device is in use and cannot be enslaved\n");
1618 if (bond_dev == slave_dev) {
1619 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1620 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1624 /* vlan challenged mutual exclusion */
1625 /* no need to lock since we're protected by rtnl_lock */
1626 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1627 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1628 if (vlan_uses_dev(bond_dev)) {
1629 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1630 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1633 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1636 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1639 if (slave_dev->features & NETIF_F_HW_ESP)
1640 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1642 /* Old ifenslave binaries are no longer supported. These can
1643 * be identified with moderate accuracy by the state of the slave:
1644 * the current ifenslave will set the interface down prior to
1645 * enslaving it; the old ifenslave will not.
1647 if (slave_dev->flags & IFF_UP) {
1648 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1649 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1653 /* set bonding device ether type by slave - bonding netdevices are
1654 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1655 * there is a need to override some of the type dependent attribs/funcs.
1657 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1658 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1660 if (!bond_has_slaves(bond)) {
1661 if (bond_dev->type != slave_dev->type) {
1662 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1663 bond_dev->type, slave_dev->type);
1665 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1667 res = notifier_to_errno(res);
1669 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1673 /* Flush unicast and multicast addresses */
1674 dev_uc_flush(bond_dev);
1675 dev_mc_flush(bond_dev);
1677 if (slave_dev->type != ARPHRD_ETHER)
1678 bond_setup_by_slave(bond_dev, slave_dev);
1680 ether_setup(bond_dev);
1681 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1684 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1687 } else if (bond_dev->type != slave_dev->type) {
1688 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1689 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1690 slave_dev->type, bond_dev->type);
1694 if (slave_dev->type == ARPHRD_INFINIBAND &&
1695 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1696 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1697 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1700 goto err_undo_flags;
1703 if (!slave_ops->ndo_set_mac_address ||
1704 slave_dev->type == ARPHRD_INFINIBAND) {
1705 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1706 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1707 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1708 if (!bond_has_slaves(bond)) {
1709 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1710 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1712 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1713 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1715 goto err_undo_flags;
1720 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1722 /* If this is the first slave, then we need to set the master's hardware
1723 * address to be the same as the slave's.
1725 if (!bond_has_slaves(bond) &&
1726 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1727 res = bond_set_dev_addr(bond->dev, slave_dev);
1729 goto err_undo_flags;
1732 new_slave = bond_alloc_slave(bond, slave_dev);
1735 goto err_undo_flags;
1738 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1739 * is set via sysfs or module option if desired.
1741 new_slave->queue_id = 0;
1743 /* Save slave's original mtu and then set it to match the bond */
1744 new_slave->original_mtu = slave_dev->mtu;
1745 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1747 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1751 /* Save slave's original ("permanent") mac address for modes
1752 * that need it, and for restoring it upon release, and then
1753 * set it to the master's address
1755 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1756 slave_dev->addr_len);
1758 if (!bond->params.fail_over_mac ||
1759 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1760 /* Set slave to master's mac address. The application already
1761 * set the master's mac address to that of the first slave
1763 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1764 ss.ss_family = slave_dev->type;
1765 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1768 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1769 goto err_restore_mtu;
1773 /* set slave flag before open to prevent IPv6 addrconf */
1774 slave_dev->flags |= IFF_SLAVE;
1776 /* open the slave since the application closed it */
1777 res = dev_open(slave_dev, extack);
1779 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1780 goto err_restore_mac;
1783 slave_dev->priv_flags |= IFF_BONDING;
1784 /* initialize slave stats */
1785 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1787 if (bond_is_lb(bond)) {
1788 /* bond_alb_init_slave() must be called before all other stages since
1789 * it might fail and we do not want to have to undo everything
1791 res = bond_alb_init_slave(bond, new_slave);
1796 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1798 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1802 prev_slave = bond_last_slave(bond);
1804 new_slave->delay = 0;
1805 new_slave->link_failure_count = 0;
1807 if (bond_update_speed_duplex(new_slave) &&
1808 bond_needs_speed_duplex(bond))
1809 new_slave->link = BOND_LINK_DOWN;
1811 new_slave->last_rx = jiffies -
1812 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1813 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1814 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1816 if (bond->params.miimon && !bond->params.use_carrier) {
1817 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1819 if ((link_reporting == -1) && !bond->params.arp_interval) {
1820 /* miimon is set but a bonded network driver
1821 * does not support ETHTOOL/MII and
1822 * arp_interval is not set. Note: if
1823 * use_carrier is enabled, we will never go
1824 * here (because netif_carrier is always
1825 * supported); thus, we don't need to change
1826 * the messages for netif_carrier.
1828 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1829 } else if (link_reporting == -1) {
1830 /* unable get link status using mii/ethtool */
1831 slave_warn(bond_dev, slave_dev, "can't get link status from slave; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n");
1835 /* check for initial state */
1836 new_slave->link = BOND_LINK_NOCHANGE;
1837 if (bond->params.miimon) {
1838 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1839 if (bond->params.updelay) {
1840 bond_set_slave_link_state(new_slave,
1842 BOND_SLAVE_NOTIFY_NOW);
1843 new_slave->delay = bond->params.updelay;
1845 bond_set_slave_link_state(new_slave,
1847 BOND_SLAVE_NOTIFY_NOW);
1850 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1851 BOND_SLAVE_NOTIFY_NOW);
1853 } else if (bond->params.arp_interval) {
1854 bond_set_slave_link_state(new_slave,
1855 (netif_carrier_ok(slave_dev) ?
1856 BOND_LINK_UP : BOND_LINK_DOWN),
1857 BOND_SLAVE_NOTIFY_NOW);
1859 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1860 BOND_SLAVE_NOTIFY_NOW);
1863 if (new_slave->link != BOND_LINK_DOWN)
1864 new_slave->last_link_up = jiffies;
1865 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1866 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1867 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1869 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1870 /* if there is a primary slave, remember it */
1871 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1872 rcu_assign_pointer(bond->primary_slave, new_slave);
1873 bond->force_primary = true;
1877 switch (BOND_MODE(bond)) {
1878 case BOND_MODE_ACTIVEBACKUP:
1879 bond_set_slave_inactive_flags(new_slave,
1880 BOND_SLAVE_NOTIFY_NOW);
1882 case BOND_MODE_8023AD:
1883 /* in 802.3ad mode, the internal mechanism
1884 * will activate the slaves in the selected
1887 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1888 /* if this is the first slave */
1890 SLAVE_AD_INFO(new_slave)->id = 1;
1891 /* Initialize AD with the number of times that the AD timer is called in 1 second
1892 * can be called only after the mac address of the bond is set
1894 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1896 SLAVE_AD_INFO(new_slave)->id =
1897 SLAVE_AD_INFO(prev_slave)->id + 1;
1900 bond_3ad_bind_slave(new_slave);
1904 bond_set_active_slave(new_slave);
1905 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1908 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1910 /* always active in trunk mode */
1911 bond_set_active_slave(new_slave);
1913 /* In trunking mode there is little meaning to curr_active_slave
1914 * anyway (it holds no special properties of the bond device),
1915 * so we can change it without calling change_active_interface()
1917 if (!rcu_access_pointer(bond->curr_active_slave) &&
1918 new_slave->link == BOND_LINK_UP)
1919 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1922 } /* switch(bond_mode) */
1924 #ifdef CONFIG_NET_POLL_CONTROLLER
1925 if (bond->dev->npinfo) {
1926 if (slave_enable_netpoll(new_slave)) {
1927 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1934 if (!(bond_dev->features & NETIF_F_LRO))
1935 dev_disable_lro(slave_dev);
1937 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1940 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1944 res = bond_master_upper_dev_link(bond, new_slave, extack);
1946 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1947 goto err_unregister;
1950 bond_lower_state_changed(new_slave);
1952 res = bond_sysfs_slave_add(new_slave);
1954 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1955 goto err_upper_unlink;
1958 /* If the mode uses primary, then the following is handled by
1959 * bond_change_active_slave().
1961 if (!bond_uses_primary(bond)) {
1962 /* set promiscuity level to new slave */
1963 if (bond_dev->flags & IFF_PROMISC) {
1964 res = dev_set_promiscuity(slave_dev, 1);
1969 /* set allmulti level to new slave */
1970 if (bond_dev->flags & IFF_ALLMULTI) {
1971 res = dev_set_allmulti(slave_dev, 1);
1973 if (bond_dev->flags & IFF_PROMISC)
1974 dev_set_promiscuity(slave_dev, -1);
1979 netif_addr_lock_bh(bond_dev);
1980 dev_mc_sync_multiple(slave_dev, bond_dev);
1981 dev_uc_sync_multiple(slave_dev, bond_dev);
1982 netif_addr_unlock_bh(bond_dev);
1984 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1985 /* add lacpdu mc addr to mc list */
1986 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1988 dev_mc_add(slave_dev, lacpdu_multicast);
1993 bond_compute_features(bond);
1994 bond_set_carrier(bond);
1996 if (bond_uses_primary(bond)) {
1998 bond_select_active_slave(bond);
1999 unblock_netpoll_tx();
2002 if (bond_mode_can_use_xmit_hash(bond))
2003 bond_update_slave_arr(bond, NULL);
2006 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2007 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2008 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2010 /* enslave is successful */
2011 bond_queue_slave_event(new_slave);
2014 /* Undo stages on error */
2016 bond_sysfs_slave_del(new_slave);
2019 bond_upper_dev_unlink(bond, new_slave);
2022 netdev_rx_handler_unregister(slave_dev);
2025 vlan_vids_del_by_dev(slave_dev, bond_dev);
2026 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2027 RCU_INIT_POINTER(bond->primary_slave, NULL);
2028 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2030 bond_change_active_slave(bond, NULL);
2031 bond_select_active_slave(bond);
2032 unblock_netpoll_tx();
2034 /* either primary_slave or curr_active_slave might've changed */
2036 slave_disable_netpoll(new_slave);
2039 if (!netif_is_bond_master(slave_dev))
2040 slave_dev->priv_flags &= ~IFF_BONDING;
2041 dev_close(slave_dev);
2044 slave_dev->flags &= ~IFF_SLAVE;
2045 if (!bond->params.fail_over_mac ||
2046 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2047 /* XXX TODO - fom follow mode needs to change master's
2048 * MAC if this slave's MAC is in use by the bond, or at
2049 * least print a warning.
2051 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2052 new_slave->dev->addr_len);
2053 ss.ss_family = slave_dev->type;
2054 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2058 dev_set_mtu(slave_dev, new_slave->original_mtu);
2061 kobject_put(&new_slave->kobj);
2064 /* Enslave of first slave has failed and we need to fix master's mac */
2065 if (!bond_has_slaves(bond)) {
2066 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2067 slave_dev->dev_addr))
2068 eth_hw_addr_random(bond_dev);
2069 if (bond_dev->type != ARPHRD_ETHER) {
2070 dev_close(bond_dev);
2071 ether_setup(bond_dev);
2072 bond_dev->flags |= IFF_MASTER;
2073 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2080 /* Try to release the slave device <slave> from the bond device <master>
2081 * It is legal to access curr_active_slave without a lock because all the function
2082 * is RTNL-locked. If "all" is true it means that the function is being called
2083 * while destroying a bond interface and all slaves are being released.
2085 * The rules for slave state should be:
2086 * for Active/Backup:
2087 * Active stays on all backups go down
2088 * for Bonded connections:
2089 * The first up interface should be left on and all others downed.
2091 static int __bond_release_one(struct net_device *bond_dev,
2092 struct net_device *slave_dev,
2093 bool all, bool unregister)
2095 struct bonding *bond = netdev_priv(bond_dev);
2096 struct slave *slave, *oldcurrent;
2097 struct sockaddr_storage ss;
2098 int old_flags = bond_dev->flags;
2099 netdev_features_t old_features = bond_dev->features;
2101 /* slave is not a slave or master is not master of this slave */
2102 if (!(slave_dev->flags & IFF_SLAVE) ||
2103 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2104 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2110 slave = bond_get_slave_by_dev(bond, slave_dev);
2112 /* not a slave of this bond */
2113 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2114 unblock_netpoll_tx();
2118 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2120 bond_sysfs_slave_del(slave);
2122 /* recompute stats just before removing the slave */
2123 bond_get_stats(bond->dev, &bond->bond_stats);
2125 bond_upper_dev_unlink(bond, slave);
2126 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2127 * for this slave anymore.
2129 netdev_rx_handler_unregister(slave_dev);
2131 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2132 bond_3ad_unbind_slave(slave);
2134 if (bond_mode_can_use_xmit_hash(bond))
2135 bond_update_slave_arr(bond, slave);
2137 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2138 bond_is_active_slave(slave) ? "active" : "backup");
2140 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2142 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2144 if (!all && (!bond->params.fail_over_mac ||
2145 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2146 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2147 bond_has_slaves(bond))
2148 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
2149 slave->perm_hwaddr);
2152 if (rtnl_dereference(bond->primary_slave) == slave)
2153 RCU_INIT_POINTER(bond->primary_slave, NULL);
2155 if (oldcurrent == slave)
2156 bond_change_active_slave(bond, NULL);
2158 if (bond_is_lb(bond)) {
2159 /* Must be called only after the slave has been
2160 * detached from the list and the curr_active_slave
2161 * has been cleared (if our_slave == old_current),
2162 * but before a new active slave is selected.
2164 bond_alb_deinit_slave(bond, slave);
2168 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2169 } else if (oldcurrent == slave) {
2170 /* Note that we hold RTNL over this sequence, so there
2171 * is no concern that another slave add/remove event
2174 bond_select_active_slave(bond);
2177 if (!bond_has_slaves(bond)) {
2178 bond_set_carrier(bond);
2179 eth_hw_addr_random(bond_dev);
2182 unblock_netpoll_tx();
2186 if (!bond_has_slaves(bond)) {
2187 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2188 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2191 bond_compute_features(bond);
2192 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2193 (old_features & NETIF_F_VLAN_CHALLENGED))
2194 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2196 vlan_vids_del_by_dev(slave_dev, bond_dev);
2198 /* If the mode uses primary, then this case was handled above by
2199 * bond_change_active_slave(..., NULL)
2201 if (!bond_uses_primary(bond)) {
2202 /* unset promiscuity level from slave
2203 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2204 * of the IFF_PROMISC flag in the bond_dev, but we need the
2205 * value of that flag before that change, as that was the value
2206 * when this slave was attached, so we cache at the start of the
2207 * function and use it here. Same goes for ALLMULTI below
2209 if (old_flags & IFF_PROMISC)
2210 dev_set_promiscuity(slave_dev, -1);
2212 /* unset allmulti level from slave */
2213 if (old_flags & IFF_ALLMULTI)
2214 dev_set_allmulti(slave_dev, -1);
2216 bond_hw_addr_flush(bond_dev, slave_dev);
2219 slave_disable_netpoll(slave);
2221 /* close slave before restoring its mac address */
2222 dev_close(slave_dev);
2224 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2225 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2226 /* restore original ("permanent") mac address */
2227 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2228 slave->dev->addr_len);
2229 ss.ss_family = slave_dev->type;
2230 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2234 __dev_set_mtu(slave_dev, slave->original_mtu);
2236 dev_set_mtu(slave_dev, slave->original_mtu);
2238 if (!netif_is_bond_master(slave_dev))
2239 slave_dev->priv_flags &= ~IFF_BONDING;
2241 kobject_put(&slave->kobj);
2246 /* A wrapper used because of ndo_del_link */
2247 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2249 return __bond_release_one(bond_dev, slave_dev, false, false);
2252 /* First release a slave and then destroy the bond if no more slaves are left.
2253 * Must be under rtnl_lock when this function is called.
2255 static int bond_release_and_destroy(struct net_device *bond_dev,
2256 struct net_device *slave_dev)
2258 struct bonding *bond = netdev_priv(bond_dev);
2261 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2262 if (ret == 0 && !bond_has_slaves(bond) &&
2263 bond_dev->reg_state != NETREG_UNREGISTERING) {
2264 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2265 netdev_info(bond_dev, "Destroying bond\n");
2266 bond_remove_proc_entry(bond);
2267 unregister_netdevice(bond_dev);
2272 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2274 struct bonding *bond = netdev_priv(bond_dev);
2275 bond_fill_ifbond(bond, info);
2278 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2280 struct bonding *bond = netdev_priv(bond_dev);
2281 struct list_head *iter;
2282 int i = 0, res = -ENODEV;
2283 struct slave *slave;
2285 bond_for_each_slave(bond, slave, iter) {
2286 if (i++ == (int)info->slave_id) {
2288 bond_fill_ifslave(slave, info);
2296 /*-------------------------------- Monitoring -------------------------------*/
2298 /* called with rcu_read_lock() */
2299 static int bond_miimon_inspect(struct bonding *bond)
2301 int link_state, commit = 0;
2302 struct list_head *iter;
2303 struct slave *slave;
2304 bool ignore_updelay;
2306 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2308 bond_for_each_slave_rcu(bond, slave, iter) {
2309 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2311 link_state = bond_check_dev_link(bond, slave->dev, 0);
2313 switch (slave->link) {
2318 bond_propose_link_state(slave, BOND_LINK_FAIL);
2320 slave->delay = bond->params.downdelay;
2322 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2324 BOND_MODE_ACTIVEBACKUP) ?
2325 (bond_is_active_slave(slave) ?
2326 "active " : "backup ") : "",
2327 bond->params.downdelay * bond->params.miimon);
2330 case BOND_LINK_FAIL:
2332 /* recovered before downdelay expired */
2333 bond_propose_link_state(slave, BOND_LINK_UP);
2334 slave->last_link_up = jiffies;
2335 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2336 (bond->params.downdelay - slave->delay) *
2337 bond->params.miimon);
2342 if (slave->delay <= 0) {
2343 bond_propose_link_state(slave, BOND_LINK_DOWN);
2351 case BOND_LINK_DOWN:
2355 bond_propose_link_state(slave, BOND_LINK_BACK);
2357 slave->delay = bond->params.updelay;
2360 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2361 ignore_updelay ? 0 :
2362 bond->params.updelay *
2363 bond->params.miimon);
2366 case BOND_LINK_BACK:
2368 bond_propose_link_state(slave, BOND_LINK_DOWN);
2369 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2370 (bond->params.updelay - slave->delay) *
2371 bond->params.miimon);
2379 if (slave->delay <= 0) {
2380 bond_propose_link_state(slave, BOND_LINK_UP);
2382 ignore_updelay = false;
2394 static void bond_miimon_link_change(struct bonding *bond,
2395 struct slave *slave,
2398 switch (BOND_MODE(bond)) {
2399 case BOND_MODE_8023AD:
2400 bond_3ad_handle_link_change(slave, link);
2404 bond_alb_handle_link_change(bond, slave, link);
2407 bond_update_slave_arr(bond, NULL);
2412 static void bond_miimon_commit(struct bonding *bond)
2414 struct list_head *iter;
2415 struct slave *slave, *primary;
2417 bond_for_each_slave(bond, slave, iter) {
2418 switch (slave->link_new_state) {
2419 case BOND_LINK_NOCHANGE:
2420 /* For 802.3ad mode, check current slave speed and
2421 * duplex again in case its port was disabled after
2422 * invalid speed/duplex reporting but recovered before
2423 * link monitoring could make a decision on the actual
2426 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2427 slave->link == BOND_LINK_UP)
2428 bond_3ad_adapter_speed_duplex_changed(slave);
2432 if (bond_update_speed_duplex(slave) &&
2433 bond_needs_speed_duplex(bond)) {
2434 slave->link = BOND_LINK_DOWN;
2435 if (net_ratelimit())
2436 slave_warn(bond->dev, slave->dev,
2437 "failed to get link speed/duplex\n");
2440 bond_set_slave_link_state(slave, BOND_LINK_UP,
2441 BOND_SLAVE_NOTIFY_NOW);
2442 slave->last_link_up = jiffies;
2444 primary = rtnl_dereference(bond->primary_slave);
2445 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2446 /* prevent it from being the active one */
2447 bond_set_backup_slave(slave);
2448 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2449 /* make it immediately active */
2450 bond_set_active_slave(slave);
2453 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2454 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2455 slave->duplex ? "full" : "half");
2457 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2459 if (!bond->curr_active_slave || slave == primary)
2464 case BOND_LINK_DOWN:
2465 if (slave->link_failure_count < UINT_MAX)
2466 slave->link_failure_count++;
2468 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2469 BOND_SLAVE_NOTIFY_NOW);
2471 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2472 BOND_MODE(bond) == BOND_MODE_8023AD)
2473 bond_set_slave_inactive_flags(slave,
2474 BOND_SLAVE_NOTIFY_NOW);
2476 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2478 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2480 if (slave == rcu_access_pointer(bond->curr_active_slave))
2486 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2487 slave->link_new_state);
2488 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2495 bond_select_active_slave(bond);
2496 unblock_netpoll_tx();
2499 bond_set_carrier(bond);
2504 * Really a wrapper that splits the mii monitor into two phases: an
2505 * inspection, then (if inspection indicates something needs to be done)
2506 * an acquisition of appropriate locks followed by a commit phase to
2507 * implement whatever link state changes are indicated.
2509 static void bond_mii_monitor(struct work_struct *work)
2511 struct bonding *bond = container_of(work, struct bonding,
2513 bool should_notify_peers = false;
2515 unsigned long delay;
2516 struct slave *slave;
2517 struct list_head *iter;
2519 delay = msecs_to_jiffies(bond->params.miimon);
2521 if (!bond_has_slaves(bond))
2525 should_notify_peers = bond_should_notify_peers(bond);
2526 commit = !!bond_miimon_inspect(bond);
2527 if (bond->send_peer_notif) {
2529 if (rtnl_trylock()) {
2530 bond->send_peer_notif--;
2538 /* Race avoidance with bond_close cancel of workqueue */
2539 if (!rtnl_trylock()) {
2541 should_notify_peers = false;
2545 bond_for_each_slave(bond, slave, iter) {
2546 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2548 bond_miimon_commit(bond);
2550 rtnl_unlock(); /* might sleep, hold no other locks */
2554 if (bond->params.miimon)
2555 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2557 if (should_notify_peers) {
2558 if (!rtnl_trylock())
2560 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2565 static int bond_upper_dev_walk(struct net_device *upper,
2566 struct netdev_nested_priv *priv)
2568 __be32 ip = *(__be32 *)priv->data;
2570 return ip == bond_confirm_addr(upper, 0, ip);
2573 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2575 struct netdev_nested_priv priv = {
2576 .data = (void *)&ip,
2580 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2584 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2591 /* We go to the (large) trouble of VLAN tagging ARP frames because
2592 * switches in VLAN mode (especially if ports are configured as
2593 * "native" to a VLAN) might not pass non-tagged frames.
2595 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2596 __be32 src_ip, struct bond_vlan_tag *tags)
2598 struct sk_buff *skb;
2599 struct bond_vlan_tag *outer_tag = tags;
2600 struct net_device *slave_dev = slave->dev;
2601 struct net_device *bond_dev = slave->bond->dev;
2603 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2604 arp_op, &dest_ip, &src_ip);
2606 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2607 NULL, slave_dev->dev_addr, NULL);
2610 net_err_ratelimited("ARP packet allocation failed\n");
2614 if (!tags || tags->vlan_proto == VLAN_N_VID)
2619 /* Go through all the tags backwards and add them to the packet */
2620 while (tags->vlan_proto != VLAN_N_VID) {
2621 if (!tags->vlan_id) {
2626 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2627 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2628 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2631 net_err_ratelimited("failed to insert inner VLAN tag\n");
2637 /* Set the outer tag */
2638 if (outer_tag->vlan_id) {
2639 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2640 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2641 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2642 outer_tag->vlan_id);
2649 /* Validate the device path between the @start_dev and the @end_dev.
2650 * The path is valid if the @end_dev is reachable through device
2652 * When the path is validated, collect any vlan information in the
2655 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2656 struct net_device *end_dev,
2659 struct bond_vlan_tag *tags;
2660 struct net_device *upper;
2661 struct list_head *iter;
2663 if (start_dev == end_dev) {
2664 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2666 return ERR_PTR(-ENOMEM);
2667 tags[level].vlan_proto = VLAN_N_VID;
2671 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2672 tags = bond_verify_device_path(upper, end_dev, level + 1);
2673 if (IS_ERR_OR_NULL(tags)) {
2678 if (is_vlan_dev(upper)) {
2679 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2680 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2689 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2692 struct bond_vlan_tag *tags;
2693 __be32 *targets = bond->params.arp_targets, addr;
2696 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2697 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2698 __func__, &targets[i]);
2701 /* Find out through which dev should the packet go */
2702 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2705 /* there's no route to target - try to send arp
2706 * probe to generate any traffic (arp_validate=0)
2708 if (bond->params.arp_validate)
2709 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2712 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2717 /* bond device itself */
2718 if (rt->dst.dev == bond->dev)
2722 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2725 if (!IS_ERR_OR_NULL(tags))
2728 /* Not our device - skip */
2729 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2730 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2736 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2738 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2743 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2747 if (!sip || !bond_has_this_ip(bond, tip)) {
2748 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2749 __func__, &sip, &tip);
2753 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2755 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2759 slave->last_rx = jiffies;
2760 slave->target_last_arp_rx[i] = jiffies;
2763 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2764 struct slave *slave)
2766 struct arphdr *arp = (struct arphdr *)skb->data;
2767 struct slave *curr_active_slave, *curr_arp_slave;
2768 unsigned char *arp_ptr;
2770 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2773 if (!slave_do_arp_validate(bond, slave)) {
2774 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2775 !slave_do_arp_validate_only(bond))
2776 slave->last_rx = jiffies;
2777 return RX_HANDLER_ANOTHER;
2778 } else if (!is_arp) {
2779 return RX_HANDLER_ANOTHER;
2782 alen = arp_hdr_len(bond->dev);
2784 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2785 __func__, skb->dev->name);
2787 if (alen > skb_headlen(skb)) {
2788 arp = kmalloc(alen, GFP_ATOMIC);
2791 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2795 if (arp->ar_hln != bond->dev->addr_len ||
2796 skb->pkt_type == PACKET_OTHERHOST ||
2797 skb->pkt_type == PACKET_LOOPBACK ||
2798 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2799 arp->ar_pro != htons(ETH_P_IP) ||
2803 arp_ptr = (unsigned char *)(arp + 1);
2804 arp_ptr += bond->dev->addr_len;
2805 memcpy(&sip, arp_ptr, 4);
2806 arp_ptr += 4 + bond->dev->addr_len;
2807 memcpy(&tip, arp_ptr, 4);
2809 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2810 __func__, slave->dev->name, bond_slave_state(slave),
2811 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2814 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2815 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2817 /* We 'trust' the received ARP enough to validate it if:
2819 * (a) the slave receiving the ARP is active (which includes the
2820 * current ARP slave, if any), or
2822 * (b) the receiving slave isn't active, but there is a currently
2823 * active slave and it received valid arp reply(s) after it became
2824 * the currently active slave, or
2826 * (c) there is an ARP slave that sent an ARP during the prior ARP
2827 * interval, and we receive an ARP reply on any slave. We accept
2828 * these because switch FDB update delays may deliver the ARP
2829 * reply to a slave other than the sender of the ARP request.
2831 * Note: for (b), backup slaves are receiving the broadcast ARP
2832 * request, not a reply. This request passes from the sending
2833 * slave through the L2 switch(es) to the receiving slave. Since
2834 * this is checking the request, sip/tip are swapped for
2837 * This is done to avoid endless looping when we can't reach the
2838 * arp_ip_target and fool ourselves with our own arp requests.
2840 if (bond_is_active_slave(slave))
2841 bond_validate_arp(bond, slave, sip, tip);
2842 else if (curr_active_slave &&
2843 time_after(slave_last_rx(bond, curr_active_slave),
2844 curr_active_slave->last_link_up))
2845 bond_validate_arp(bond, slave, tip, sip);
2846 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2847 bond_time_in_interval(bond,
2848 dev_trans_start(curr_arp_slave->dev), 1))
2849 bond_validate_arp(bond, slave, sip, tip);
2852 if (arp != (struct arphdr *)skb->data)
2854 return RX_HANDLER_ANOTHER;
2857 /* function to verify if we're in the arp_interval timeslice, returns true if
2858 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2859 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2861 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2864 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2866 return time_in_range(jiffies,
2867 last_act - delta_in_ticks,
2868 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2871 /* This function is called regularly to monitor each slave's link
2872 * ensuring that traffic is being sent and received when arp monitoring
2873 * is used in load-balancing mode. if the adapter has been dormant, then an
2874 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2875 * arp monitoring in active backup mode.
2877 static void bond_loadbalance_arp_mon(struct bonding *bond)
2879 struct slave *slave, *oldcurrent;
2880 struct list_head *iter;
2881 int do_failover = 0, slave_state_changed = 0;
2883 if (!bond_has_slaves(bond))
2888 oldcurrent = rcu_dereference(bond->curr_active_slave);
2889 /* see if any of the previous devices are up now (i.e. they have
2890 * xmt and rcv traffic). the curr_active_slave does not come into
2891 * the picture unless it is null. also, slave->last_link_up is not
2892 * needed here because we send an arp on each slave and give a slave
2893 * as long as it needs to get the tx/rx within the delta.
2894 * TODO: what about up/down delay in arp mode? it wasn't here before
2897 bond_for_each_slave_rcu(bond, slave, iter) {
2898 unsigned long trans_start = dev_trans_start(slave->dev);
2900 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2902 if (slave->link != BOND_LINK_UP) {
2903 if (bond_time_in_interval(bond, trans_start, 1) &&
2904 bond_time_in_interval(bond, slave->last_rx, 1)) {
2906 bond_propose_link_state(slave, BOND_LINK_UP);
2907 slave_state_changed = 1;
2909 /* primary_slave has no meaning in round-robin
2910 * mode. the window of a slave being up and
2911 * curr_active_slave being null after enslaving
2915 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2918 slave_info(bond->dev, slave->dev, "interface is now up\n");
2922 /* slave->link == BOND_LINK_UP */
2924 /* not all switches will respond to an arp request
2925 * when the source ip is 0, so don't take the link down
2926 * if we don't know our ip yet
2928 if (!bond_time_in_interval(bond, trans_start, 2) ||
2929 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2931 bond_propose_link_state(slave, BOND_LINK_DOWN);
2932 slave_state_changed = 1;
2934 if (slave->link_failure_count < UINT_MAX)
2935 slave->link_failure_count++;
2937 slave_info(bond->dev, slave->dev, "interface is now down\n");
2939 if (slave == oldcurrent)
2944 /* note: if switch is in round-robin mode, all links
2945 * must tx arp to ensure all links rx an arp - otherwise
2946 * links may oscillate or not come up at all; if switch is
2947 * in something like xor mode, there is nothing we can
2948 * do - all replies will be rx'ed on same link causing slaves
2949 * to be unstable during low/no traffic periods
2951 if (bond_slave_is_up(slave))
2952 bond_arp_send_all(bond, slave);
2957 if (do_failover || slave_state_changed) {
2958 if (!rtnl_trylock())
2961 bond_for_each_slave(bond, slave, iter) {
2962 if (slave->link_new_state != BOND_LINK_NOCHANGE)
2963 slave->link = slave->link_new_state;
2966 if (slave_state_changed) {
2967 bond_slave_state_change(bond);
2968 if (BOND_MODE(bond) == BOND_MODE_XOR)
2969 bond_update_slave_arr(bond, NULL);
2973 bond_select_active_slave(bond);
2974 unblock_netpoll_tx();
2980 if (bond->params.arp_interval)
2981 queue_delayed_work(bond->wq, &bond->arp_work,
2982 msecs_to_jiffies(bond->params.arp_interval));
2985 /* Called to inspect slaves for active-backup mode ARP monitor link state
2986 * changes. Sets proposed link state in slaves to specify what action
2987 * should take place for the slave. Returns 0 if no changes are found, >0
2988 * if changes to link states must be committed.
2990 * Called with rcu_read_lock held.
2992 static int bond_ab_arp_inspect(struct bonding *bond)
2994 unsigned long trans_start, last_rx;
2995 struct list_head *iter;
2996 struct slave *slave;
2999 bond_for_each_slave_rcu(bond, slave, iter) {
3000 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3001 last_rx = slave_last_rx(bond, slave);
3003 if (slave->link != BOND_LINK_UP) {
3004 if (bond_time_in_interval(bond, last_rx, 1)) {
3005 bond_propose_link_state(slave, BOND_LINK_UP);
3007 } else if (slave->link == BOND_LINK_BACK) {
3008 bond_propose_link_state(slave, BOND_LINK_FAIL);
3014 /* Give slaves 2*delta after being enslaved or made
3015 * active. This avoids bouncing, as the last receive
3016 * times need a full ARP monitor cycle to be updated.
3018 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3021 /* Backup slave is down if:
3022 * - No current_arp_slave AND
3023 * - more than 3*delta since last receive AND
3024 * - the bond has an IP address
3026 * Note: a non-null current_arp_slave indicates
3027 * the curr_active_slave went down and we are
3028 * searching for a new one; under this condition
3029 * we only take the curr_active_slave down - this
3030 * gives each slave a chance to tx/rx traffic
3031 * before being taken out
3033 if (!bond_is_active_slave(slave) &&
3034 !rcu_access_pointer(bond->current_arp_slave) &&
3035 !bond_time_in_interval(bond, last_rx, 3)) {
3036 bond_propose_link_state(slave, BOND_LINK_DOWN);
3040 /* Active slave is down if:
3041 * - more than 2*delta since transmitting OR
3042 * - (more than 2*delta since receive AND
3043 * the bond has an IP address)
3045 trans_start = dev_trans_start(slave->dev);
3046 if (bond_is_active_slave(slave) &&
3047 (!bond_time_in_interval(bond, trans_start, 2) ||
3048 !bond_time_in_interval(bond, last_rx, 2))) {
3049 bond_propose_link_state(slave, BOND_LINK_DOWN);
3057 /* Called to commit link state changes noted by inspection step of
3058 * active-backup mode ARP monitor.
3060 * Called with RTNL hold.
3062 static void bond_ab_arp_commit(struct bonding *bond)
3064 unsigned long trans_start;
3065 struct list_head *iter;
3066 struct slave *slave;
3068 bond_for_each_slave(bond, slave, iter) {
3069 switch (slave->link_new_state) {
3070 case BOND_LINK_NOCHANGE:
3074 trans_start = dev_trans_start(slave->dev);
3075 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3076 (!rtnl_dereference(bond->curr_active_slave) &&
3077 bond_time_in_interval(bond, trans_start, 1))) {
3078 struct slave *current_arp_slave;
3080 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3081 bond_set_slave_link_state(slave, BOND_LINK_UP,
3082 BOND_SLAVE_NOTIFY_NOW);
3083 if (current_arp_slave) {
3084 bond_set_slave_inactive_flags(
3086 BOND_SLAVE_NOTIFY_NOW);
3087 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3090 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3092 if (!rtnl_dereference(bond->curr_active_slave) ||
3093 slave == rtnl_dereference(bond->primary_slave))
3100 case BOND_LINK_DOWN:
3101 if (slave->link_failure_count < UINT_MAX)
3102 slave->link_failure_count++;
3104 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3105 BOND_SLAVE_NOTIFY_NOW);
3106 bond_set_slave_inactive_flags(slave,
3107 BOND_SLAVE_NOTIFY_NOW);
3109 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3111 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3112 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3118 case BOND_LINK_FAIL:
3119 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3120 BOND_SLAVE_NOTIFY_NOW);
3121 bond_set_slave_inactive_flags(slave,
3122 BOND_SLAVE_NOTIFY_NOW);
3124 /* A slave has just been enslaved and has become
3125 * the current active slave.
3127 if (rtnl_dereference(bond->curr_active_slave))
3128 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3132 slave_err(bond->dev, slave->dev,
3133 "impossible: link_new_state %d on slave\n",
3134 slave->link_new_state);
3140 bond_select_active_slave(bond);
3141 unblock_netpoll_tx();
3144 bond_set_carrier(bond);
3147 /* Send ARP probes for active-backup mode ARP monitor.
3149 * Called with rcu_read_lock held.
3151 static bool bond_ab_arp_probe(struct bonding *bond)
3153 struct slave *slave, *before = NULL, *new_slave = NULL,
3154 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3155 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3156 struct list_head *iter;
3158 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3160 if (curr_arp_slave && curr_active_slave)
3161 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3162 curr_arp_slave->dev->name,
3163 curr_active_slave->dev->name);
3165 if (curr_active_slave) {
3166 bond_arp_send_all(bond, curr_active_slave);
3167 return should_notify_rtnl;
3170 /* if we don't have a curr_active_slave, search for the next available
3171 * backup slave from the current_arp_slave and make it the candidate
3172 * for becoming the curr_active_slave
3175 if (!curr_arp_slave) {
3176 curr_arp_slave = bond_first_slave_rcu(bond);
3177 if (!curr_arp_slave)
3178 return should_notify_rtnl;
3181 bond_for_each_slave_rcu(bond, slave, iter) {
3182 if (!found && !before && bond_slave_is_up(slave))
3185 if (found && !new_slave && bond_slave_is_up(slave))
3187 /* if the link state is up at this point, we
3188 * mark it down - this can happen if we have
3189 * simultaneous link failures and
3190 * reselect_active_interface doesn't make this
3191 * one the current slave so it is still marked
3192 * up when it is actually down
3194 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3195 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3196 BOND_SLAVE_NOTIFY_LATER);
3197 if (slave->link_failure_count < UINT_MAX)
3198 slave->link_failure_count++;
3200 bond_set_slave_inactive_flags(slave,
3201 BOND_SLAVE_NOTIFY_LATER);
3203 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3205 if (slave == curr_arp_slave)
3209 if (!new_slave && before)
3215 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3216 BOND_SLAVE_NOTIFY_LATER);
3217 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3218 bond_arp_send_all(bond, new_slave);
3219 new_slave->last_link_up = jiffies;
3220 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3223 bond_for_each_slave_rcu(bond, slave, iter) {
3224 if (slave->should_notify || slave->should_notify_link) {
3225 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3229 return should_notify_rtnl;
3232 static void bond_activebackup_arp_mon(struct bonding *bond)
3234 bool should_notify_peers = false;
3235 bool should_notify_rtnl = false;
3238 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3240 if (!bond_has_slaves(bond))
3245 should_notify_peers = bond_should_notify_peers(bond);
3247 if (bond_ab_arp_inspect(bond)) {
3250 /* Race avoidance with bond_close flush of workqueue */
3251 if (!rtnl_trylock()) {
3253 should_notify_peers = false;
3257 bond_ab_arp_commit(bond);
3263 should_notify_rtnl = bond_ab_arp_probe(bond);
3267 if (bond->params.arp_interval)
3268 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3270 if (should_notify_peers || should_notify_rtnl) {
3271 if (!rtnl_trylock())
3274 if (should_notify_peers)
3275 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3277 if (should_notify_rtnl) {
3278 bond_slave_state_notify(bond);
3279 bond_slave_link_notify(bond);
3286 static void bond_arp_monitor(struct work_struct *work)
3288 struct bonding *bond = container_of(work, struct bonding,
3291 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3292 bond_activebackup_arp_mon(bond);
3294 bond_loadbalance_arp_mon(bond);
3297 /*-------------------------- netdev event handling --------------------------*/
3299 /* Change device name */
3300 static int bond_event_changename(struct bonding *bond)
3302 bond_remove_proc_entry(bond);
3303 bond_create_proc_entry(bond);
3305 bond_debug_reregister(bond);
3310 static int bond_master_netdev_event(unsigned long event,
3311 struct net_device *bond_dev)
3313 struct bonding *event_bond = netdev_priv(bond_dev);
3315 netdev_dbg(bond_dev, "%s called\n", __func__);
3318 case NETDEV_CHANGENAME:
3319 return bond_event_changename(event_bond);
3320 case NETDEV_UNREGISTER:
3321 bond_remove_proc_entry(event_bond);
3323 case NETDEV_REGISTER:
3324 bond_create_proc_entry(event_bond);
3333 static int bond_slave_netdev_event(unsigned long event,
3334 struct net_device *slave_dev)
3336 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3337 struct bonding *bond;
3338 struct net_device *bond_dev;
3340 /* A netdev event can be generated while enslaving a device
3341 * before netdev_rx_handler_register is called in which case
3342 * slave will be NULL
3345 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3349 bond_dev = slave->bond->dev;
3351 primary = rtnl_dereference(bond->primary_slave);
3353 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3356 case NETDEV_UNREGISTER:
3357 if (bond_dev->type != ARPHRD_ETHER)
3358 bond_release_and_destroy(bond_dev, slave_dev);
3360 __bond_release_one(bond_dev, slave_dev, false, true);
3364 /* For 802.3ad mode only:
3365 * Getting invalid Speed/Duplex values here will put slave
3366 * in weird state. Mark it as link-fail if the link was
3367 * previously up or link-down if it hasn't yet come up, and
3368 * let link-monitoring (miimon) set it right when correct
3369 * speeds/duplex are available.
3371 if (bond_update_speed_duplex(slave) &&
3372 BOND_MODE(bond) == BOND_MODE_8023AD) {
3373 if (slave->last_link_up)
3374 slave->link = BOND_LINK_FAIL;
3376 slave->link = BOND_LINK_DOWN;
3379 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3380 bond_3ad_adapter_speed_duplex_changed(slave);
3383 /* Refresh slave-array if applicable!
3384 * If the setup does not use miimon or arpmon (mode-specific!),
3385 * then these events will not cause the slave-array to be
3386 * refreshed. This will cause xmit to use a slave that is not
3387 * usable. Avoid such situation by refeshing the array at these
3388 * events. If these (miimon/arpmon) parameters are configured
3389 * then array gets refreshed twice and that should be fine!
3391 if (bond_mode_can_use_xmit_hash(bond))
3392 bond_update_slave_arr(bond, NULL);
3394 case NETDEV_CHANGEMTU:
3395 /* TODO: Should slaves be allowed to
3396 * independently alter their MTU? For
3397 * an active-backup bond, slaves need
3398 * not be the same type of device, so
3399 * MTUs may vary. For other modes,
3400 * slaves arguably should have the
3401 * same MTUs. To do this, we'd need to
3402 * take over the slave's change_mtu
3403 * function for the duration of their
3407 case NETDEV_CHANGENAME:
3408 /* we don't care if we don't have primary set */
3409 if (!bond_uses_primary(bond) ||
3410 !bond->params.primary[0])
3413 if (slave == primary) {
3414 /* slave's name changed - he's no longer primary */
3415 RCU_INIT_POINTER(bond->primary_slave, NULL);
3416 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3417 /* we have a new primary slave */
3418 rcu_assign_pointer(bond->primary_slave, slave);
3419 } else { /* we didn't change primary - exit */
3423 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3424 primary ? slave_dev->name : "none");
3427 bond_select_active_slave(bond);
3428 unblock_netpoll_tx();
3430 case NETDEV_FEAT_CHANGE:
3431 bond_compute_features(bond);
3433 case NETDEV_RESEND_IGMP:
3434 /* Propagate to master device */
3435 call_netdevice_notifiers(event, slave->bond->dev);
3444 /* bond_netdev_event: handle netdev notifier chain events.
3446 * This function receives events for the netdev chain. The caller (an
3447 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3448 * locks for us to safely manipulate the slave devices (RTNL lock,
3451 static int bond_netdev_event(struct notifier_block *this,
3452 unsigned long event, void *ptr)
3454 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3456 netdev_dbg(event_dev, "%s received %s\n",
3457 __func__, netdev_cmd_to_name(event));
3459 if (!(event_dev->priv_flags & IFF_BONDING))
3462 if (event_dev->flags & IFF_MASTER) {
3465 ret = bond_master_netdev_event(event, event_dev);
3466 if (ret != NOTIFY_DONE)
3470 if (event_dev->flags & IFF_SLAVE)
3471 return bond_slave_netdev_event(event, event_dev);
3476 static struct notifier_block bond_netdev_notifier = {
3477 .notifier_call = bond_netdev_event,
3480 /*---------------------------- Hashing Policies -----------------------------*/
3482 /* L2 hash helper */
3483 static inline u32 bond_eth_hash(struct sk_buff *skb)
3485 struct ethhdr *ep, hdr_tmp;
3487 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3489 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3493 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3494 int *noff, int *proto, bool l34)
3496 const struct ipv6hdr *iph6;
3497 const struct iphdr *iph;
3499 if (skb->protocol == htons(ETH_P_IP)) {
3500 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3502 iph = (const struct iphdr *)(skb->data + *noff);
3503 iph_to_flow_copy_v4addrs(fk, iph);
3504 *noff += iph->ihl << 2;
3505 if (!ip_is_fragment(iph))
3506 *proto = iph->protocol;
3507 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3508 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3510 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3511 iph_to_flow_copy_v6addrs(fk, iph6);
3512 *noff += sizeof(*iph6);
3513 *proto = iph6->nexthdr;
3518 if (l34 && *proto >= 0)
3519 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3524 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb)
3526 struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
3527 u32 srcmac_vendor = 0, srcmac_dev = 0;
3531 for (i = 0; i < 3; i++)
3532 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3534 for (i = 3; i < ETH_ALEN; i++)
3535 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3537 if (!skb_vlan_tag_present(skb))
3538 return srcmac_vendor ^ srcmac_dev;
3540 vlan = skb_vlan_tag_get(skb);
3542 return vlan ^ srcmac_vendor ^ srcmac_dev;
3545 /* Extract the appropriate headers based on bond's xmit policy */
3546 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3547 struct flow_keys *fk)
3549 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3550 int noff, proto = -1;
3552 switch (bond->params.xmit_policy) {
3553 case BOND_XMIT_POLICY_ENCAP23:
3554 case BOND_XMIT_POLICY_ENCAP34:
3555 memset(fk, 0, sizeof(*fk));
3556 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3557 fk, NULL, 0, 0, 0, 0);
3562 fk->ports.ports = 0;
3563 memset(&fk->icmp, 0, sizeof(fk->icmp));
3564 noff = skb_network_offset(skb);
3565 if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3568 /* ICMP error packets contains at least 8 bytes of the header
3569 * of the packet which generated the error. Use this information
3570 * to correlate ICMP error packets within the same flow which
3571 * generated the error.
3573 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3574 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3575 skb_transport_offset(skb),
3577 if (proto == IPPROTO_ICMP) {
3578 if (!icmp_is_err(fk->icmp.type))
3581 noff += sizeof(struct icmphdr);
3582 } else if (proto == IPPROTO_ICMPV6) {
3583 if (!icmpv6_is_err(fk->icmp.type))
3586 noff += sizeof(struct icmp6hdr);
3588 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3594 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow)
3596 hash ^= (__force u32)flow_get_u32_dst(flow) ^
3597 (__force u32)flow_get_u32_src(flow);
3598 hash ^= (hash >> 16);
3599 hash ^= (hash >> 8);
3600 /* discard lowest hash bit to deal with the common even ports pattern */
3605 * bond_xmit_hash - generate a hash value based on the xmit policy
3606 * @bond: bonding device
3607 * @skb: buffer to use for headers
3609 * This function will extract the necessary headers from the skb buffer and use
3610 * them to generate a hash based on the xmit_policy set in the bonding device
3612 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3614 struct flow_keys flow;
3617 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3621 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
3622 return bond_vlan_srcmac_hash(skb);
3624 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3625 !bond_flow_dissect(bond, skb, &flow))
3626 return bond_eth_hash(skb);
3628 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3629 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3630 hash = bond_eth_hash(skb);
3633 memcpy(&hash, &flow.icmp, sizeof(hash));
3635 memcpy(&hash, &flow.ports.ports, sizeof(hash));
3638 return bond_ip_hash(hash, &flow);
3641 /*-------------------------- Device entry points ----------------------------*/
3643 void bond_work_init_all(struct bonding *bond)
3645 INIT_DELAYED_WORK(&bond->mcast_work,
3646 bond_resend_igmp_join_requests_delayed);
3647 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3648 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3649 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3650 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3651 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3654 static void bond_work_cancel_all(struct bonding *bond)
3656 cancel_delayed_work_sync(&bond->mii_work);
3657 cancel_delayed_work_sync(&bond->arp_work);
3658 cancel_delayed_work_sync(&bond->alb_work);
3659 cancel_delayed_work_sync(&bond->ad_work);
3660 cancel_delayed_work_sync(&bond->mcast_work);
3661 cancel_delayed_work_sync(&bond->slave_arr_work);
3664 static int bond_open(struct net_device *bond_dev)
3666 struct bonding *bond = netdev_priv(bond_dev);
3667 struct list_head *iter;
3668 struct slave *slave;
3670 /* reset slave->backup and slave->inactive */
3671 if (bond_has_slaves(bond)) {
3672 bond_for_each_slave(bond, slave, iter) {
3673 if (bond_uses_primary(bond) &&
3674 slave != rcu_access_pointer(bond->curr_active_slave)) {
3675 bond_set_slave_inactive_flags(slave,
3676 BOND_SLAVE_NOTIFY_NOW);
3677 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3678 bond_set_slave_active_flags(slave,
3679 BOND_SLAVE_NOTIFY_NOW);
3684 if (bond_is_lb(bond)) {
3685 /* bond_alb_initialize must be called before the timer
3688 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3690 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3691 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3694 if (bond->params.miimon) /* link check interval, in milliseconds. */
3695 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3697 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3698 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3699 bond->recv_probe = bond_arp_rcv;
3702 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3703 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3704 /* register to receive LACPDUs */
3705 bond->recv_probe = bond_3ad_lacpdu_recv;
3706 bond_3ad_initiate_agg_selection(bond, 1);
3709 if (bond_mode_can_use_xmit_hash(bond))
3710 bond_update_slave_arr(bond, NULL);
3715 static int bond_close(struct net_device *bond_dev)
3717 struct bonding *bond = netdev_priv(bond_dev);
3719 bond_work_cancel_all(bond);
3720 bond->send_peer_notif = 0;
3721 if (bond_is_lb(bond))
3722 bond_alb_deinitialize(bond);
3723 bond->recv_probe = NULL;
3728 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3729 * that some drivers can provide 32bit values only.
3731 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3732 const struct rtnl_link_stats64 *_new,
3733 const struct rtnl_link_stats64 *_old)
3735 const u64 *new = (const u64 *)_new;
3736 const u64 *old = (const u64 *)_old;
3737 u64 *res = (u64 *)_res;
3740 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3743 s64 delta = nv - ov;
3745 /* detects if this particular field is 32bit only */
3746 if (((nv | ov) >> 32) == 0)
3747 delta = (s64)(s32)((u32)nv - (u32)ov);
3749 /* filter anomalies, some drivers reset their stats
3750 * at down/up events.
3757 #ifdef CONFIG_LOCKDEP
3758 static int bond_get_lowest_level_rcu(struct net_device *dev)
3760 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3761 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3762 int cur = 0, max = 0;
3765 iter = &dev->adj_list.lower;
3770 ldev = netdev_next_lower_dev_rcu(now, &iter);
3775 niter = &ldev->adj_list.lower;
3776 dev_stack[cur] = now;
3777 iter_stack[cur++] = iter;
3786 next = dev_stack[--cur];
3787 niter = iter_stack[cur];
3798 static void bond_get_stats(struct net_device *bond_dev,
3799 struct rtnl_link_stats64 *stats)
3801 struct bonding *bond = netdev_priv(bond_dev);
3802 struct rtnl_link_stats64 temp;
3803 struct list_head *iter;
3804 struct slave *slave;
3809 #ifdef CONFIG_LOCKDEP
3810 nest_level = bond_get_lowest_level_rcu(bond_dev);
3813 spin_lock_nested(&bond->stats_lock, nest_level);
3814 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3816 bond_for_each_slave_rcu(bond, slave, iter) {
3817 const struct rtnl_link_stats64 *new =
3818 dev_get_stats(slave->dev, &temp);
3820 bond_fold_stats(stats, new, &slave->slave_stats);
3822 /* save off the slave stats for the next run */
3823 memcpy(&slave->slave_stats, new, sizeof(*new));
3826 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3827 spin_unlock(&bond->stats_lock);
3831 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3833 struct bonding *bond = netdev_priv(bond_dev);
3834 struct net_device *slave_dev = NULL;
3835 struct ifbond k_binfo;
3836 struct ifbond __user *u_binfo = NULL;
3837 struct ifslave k_sinfo;
3838 struct ifslave __user *u_sinfo = NULL;
3839 struct mii_ioctl_data *mii = NULL;
3840 struct bond_opt_value newval;
3844 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3855 /* We do this again just in case we were called by SIOCGMIIREG
3856 * instead of SIOCGMIIPHY.
3862 if (mii->reg_num == 1) {
3864 if (netif_carrier_ok(bond->dev))
3865 mii->val_out = BMSR_LSTATUS;
3869 case BOND_INFO_QUERY_OLD:
3870 case SIOCBONDINFOQUERY:
3871 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3873 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3876 bond_info_query(bond_dev, &k_binfo);
3877 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3881 case BOND_SLAVE_INFO_QUERY_OLD:
3882 case SIOCBONDSLAVEINFOQUERY:
3883 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3885 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3888 res = bond_slave_info_query(bond_dev, &k_sinfo);
3890 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3898 net = dev_net(bond_dev);
3900 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3903 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3905 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3911 case BOND_ENSLAVE_OLD:
3912 case SIOCBONDENSLAVE:
3913 res = bond_enslave(bond_dev, slave_dev, NULL);
3915 case BOND_RELEASE_OLD:
3916 case SIOCBONDRELEASE:
3917 res = bond_release(bond_dev, slave_dev);
3919 case BOND_SETHWADDR_OLD:
3920 case SIOCBONDSETHWADDR:
3921 res = bond_set_dev_addr(bond_dev, slave_dev);
3923 case BOND_CHANGE_ACTIVE_OLD:
3924 case SIOCBONDCHANGEACTIVE:
3925 bond_opt_initstr(&newval, slave_dev->name);
3926 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3936 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3938 struct bonding *bond = netdev_priv(bond_dev);
3940 if (change & IFF_PROMISC)
3941 bond_set_promiscuity(bond,
3942 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3944 if (change & IFF_ALLMULTI)
3945 bond_set_allmulti(bond,
3946 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3949 static void bond_set_rx_mode(struct net_device *bond_dev)
3951 struct bonding *bond = netdev_priv(bond_dev);
3952 struct list_head *iter;
3953 struct slave *slave;
3956 if (bond_uses_primary(bond)) {
3957 slave = rcu_dereference(bond->curr_active_slave);
3959 dev_uc_sync(slave->dev, bond_dev);
3960 dev_mc_sync(slave->dev, bond_dev);
3963 bond_for_each_slave_rcu(bond, slave, iter) {
3964 dev_uc_sync_multiple(slave->dev, bond_dev);
3965 dev_mc_sync_multiple(slave->dev, bond_dev);
3971 static int bond_neigh_init(struct neighbour *n)
3973 struct bonding *bond = netdev_priv(n->dev);
3974 const struct net_device_ops *slave_ops;
3975 struct neigh_parms parms;
3976 struct slave *slave;
3980 slave = bond_first_slave_rcu(bond);
3983 slave_ops = slave->dev->netdev_ops;
3984 if (!slave_ops->ndo_neigh_setup)
3987 /* TODO: find another way [1] to implement this.
3988 * Passing a zeroed structure is fragile,
3989 * but at least we do not pass garbage.
3991 * [1] One way would be that ndo_neigh_setup() never touch
3992 * struct neigh_parms, but propagate the new neigh_setup()
3993 * back to ___neigh_create() / neigh_parms_alloc()
3995 memset(&parms, 0, sizeof(parms));
3996 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4001 if (parms.neigh_setup)
4002 ret = parms.neigh_setup(n);
4008 /* The bonding ndo_neigh_setup is called at init time beofre any
4009 * slave exists. So we must declare proxy setup function which will
4010 * be used at run time to resolve the actual slave neigh param setup.
4012 * It's also called by master devices (such as vlans) to setup their
4013 * underlying devices. In that case - do nothing, we're already set up from
4016 static int bond_neigh_setup(struct net_device *dev,
4017 struct neigh_parms *parms)
4019 /* modify only our neigh_parms */
4020 if (parms->dev == dev)
4021 parms->neigh_setup = bond_neigh_init;
4026 /* Change the MTU of all of a master's slaves to match the master */
4027 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4029 struct bonding *bond = netdev_priv(bond_dev);
4030 struct slave *slave, *rollback_slave;
4031 struct list_head *iter;
4034 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4036 bond_for_each_slave(bond, slave, iter) {
4037 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4038 slave, slave->dev->netdev_ops->ndo_change_mtu);
4040 res = dev_set_mtu(slave->dev, new_mtu);
4043 /* If we failed to set the slave's mtu to the new value
4044 * we must abort the operation even in ACTIVE_BACKUP
4045 * mode, because if we allow the backup slaves to have
4046 * different mtu values than the active slave we'll
4047 * need to change their mtu when doing a failover. That
4048 * means changing their mtu from timer context, which
4049 * is probably not a good idea.
4051 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4057 bond_dev->mtu = new_mtu;
4062 /* unwind from head to the slave that failed */
4063 bond_for_each_slave(bond, rollback_slave, iter) {
4066 if (rollback_slave == slave)
4069 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4071 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4078 /* Change HW address
4080 * Note that many devices must be down to change the HW address, and
4081 * downing the master releases all slaves. We can make bonds full of
4082 * bonding devices to test this, however.
4084 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4086 struct bonding *bond = netdev_priv(bond_dev);
4087 struct slave *slave, *rollback_slave;
4088 struct sockaddr_storage *ss = addr, tmp_ss;
4089 struct list_head *iter;
4092 if (BOND_MODE(bond) == BOND_MODE_ALB)
4093 return bond_alb_set_mac_address(bond_dev, addr);
4096 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4098 /* If fail_over_mac is enabled, do nothing and return success.
4099 * Returning an error causes ifenslave to fail.
4101 if (bond->params.fail_over_mac &&
4102 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4105 if (!is_valid_ether_addr(ss->__data))
4106 return -EADDRNOTAVAIL;
4108 bond_for_each_slave(bond, slave, iter) {
4109 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4111 res = dev_set_mac_address(slave->dev, addr, NULL);
4113 /* TODO: consider downing the slave
4115 * User should expect communications
4116 * breakage anyway until ARP finish
4119 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4126 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4130 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4131 tmp_ss.ss_family = bond_dev->type;
4133 /* unwind from head to the slave that failed */
4134 bond_for_each_slave(bond, rollback_slave, iter) {
4137 if (rollback_slave == slave)
4140 tmp_res = dev_set_mac_address(rollback_slave->dev,
4141 (struct sockaddr *)&tmp_ss, NULL);
4143 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4152 * bond_get_slave_by_id - get xmit slave with slave_id
4153 * @bond: bonding device that is transmitting
4154 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4156 * This function tries to get slave with slave_id but in case
4157 * it fails, it tries to find the first available slave for transmission.
4159 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4162 struct list_head *iter;
4163 struct slave *slave;
4166 /* Here we start from the slave with slave_id */
4167 bond_for_each_slave_rcu(bond, slave, iter) {
4169 if (bond_slave_can_tx(slave))
4174 /* Here we start from the first slave up to slave_id */
4176 bond_for_each_slave_rcu(bond, slave, iter) {
4179 if (bond_slave_can_tx(slave))
4182 /* no slave that can tx has been found */
4187 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4188 * @bond: bonding device to use
4190 * Based on the value of the bonding device's packets_per_slave parameter
4191 * this function generates a slave id, which is usually used as the next
4192 * slave to transmit through.
4194 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4197 struct reciprocal_value reciprocal_packets_per_slave;
4198 int packets_per_slave = bond->params.packets_per_slave;
4200 switch (packets_per_slave) {
4202 slave_id = prandom_u32();
4205 slave_id = bond->rr_tx_counter;
4208 reciprocal_packets_per_slave =
4209 bond->params.reciprocal_packets_per_slave;
4210 slave_id = reciprocal_divide(bond->rr_tx_counter,
4211 reciprocal_packets_per_slave);
4214 bond->rr_tx_counter++;
4219 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4220 struct sk_buff *skb)
4222 struct slave *slave;
4226 /* Start with the curr_active_slave that joined the bond as the
4227 * default for sending IGMP traffic. For failover purposes one
4228 * needs to maintain some consistency for the interface that will
4229 * send the join/membership reports. The curr_active_slave found
4230 * will send all of this type of traffic.
4232 if (skb->protocol == htons(ETH_P_IP)) {
4233 int noff = skb_network_offset(skb);
4236 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4240 if (iph->protocol == IPPROTO_IGMP) {
4241 slave = rcu_dereference(bond->curr_active_slave);
4244 return bond_get_slave_by_id(bond, 0);
4249 slave_cnt = READ_ONCE(bond->slave_cnt);
4250 if (likely(slave_cnt)) {
4251 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4252 return bond_get_slave_by_id(bond, slave_id);
4257 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4258 struct net_device *bond_dev)
4260 struct bonding *bond = netdev_priv(bond_dev);
4261 struct slave *slave;
4263 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4265 return bond_dev_queue_xmit(bond, skb, slave->dev);
4267 return bond_tx_drop(bond_dev, skb);
4270 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4271 struct sk_buff *skb)
4273 return rcu_dereference(bond->curr_active_slave);
4276 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4277 * the bond has a usable interface.
4279 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4280 struct net_device *bond_dev)
4282 struct bonding *bond = netdev_priv(bond_dev);
4283 struct slave *slave;
4285 slave = bond_xmit_activebackup_slave_get(bond, skb);
4287 return bond_dev_queue_xmit(bond, skb, slave->dev);
4289 return bond_tx_drop(bond_dev, skb);
4292 /* Use this to update slave_array when (a) it's not appropriate to update
4293 * slave_array right away (note that update_slave_array() may sleep)
4294 * and / or (b) RTNL is not held.
4296 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4298 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4301 /* Slave array work handler. Holds only RTNL */
4302 static void bond_slave_arr_handler(struct work_struct *work)
4304 struct bonding *bond = container_of(work, struct bonding,
4305 slave_arr_work.work);
4308 if (!rtnl_trylock())
4311 ret = bond_update_slave_arr(bond, NULL);
4314 pr_warn_ratelimited("Failed to update slave array from WT\n");
4320 bond_slave_arr_work_rearm(bond, 1);
4323 static void bond_skip_slave(struct bond_up_slave *slaves,
4324 struct slave *skipslave)
4328 /* Rare situation where caller has asked to skip a specific
4329 * slave but allocation failed (most likely!). BTW this is
4330 * only possible when the call is initiated from
4331 * __bond_release_one(). In this situation; overwrite the
4332 * skipslave entry in the array with the last entry from the
4333 * array to avoid a situation where the xmit path may choose
4334 * this to-be-skipped slave to send a packet out.
4336 for (idx = 0; slaves && idx < slaves->count; idx++) {
4337 if (skipslave == slaves->arr[idx]) {
4339 slaves->arr[slaves->count - 1];
4346 static void bond_set_slave_arr(struct bonding *bond,
4347 struct bond_up_slave *usable_slaves,
4348 struct bond_up_slave *all_slaves)
4350 struct bond_up_slave *usable, *all;
4352 usable = rtnl_dereference(bond->usable_slaves);
4353 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4354 kfree_rcu(usable, rcu);
4356 all = rtnl_dereference(bond->all_slaves);
4357 rcu_assign_pointer(bond->all_slaves, all_slaves);
4358 kfree_rcu(all, rcu);
4361 static void bond_reset_slave_arr(struct bonding *bond)
4363 struct bond_up_slave *usable, *all;
4365 usable = rtnl_dereference(bond->usable_slaves);
4367 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4368 kfree_rcu(usable, rcu);
4371 all = rtnl_dereference(bond->all_slaves);
4373 RCU_INIT_POINTER(bond->all_slaves, NULL);
4374 kfree_rcu(all, rcu);
4378 /* Build the usable slaves array in control path for modes that use xmit-hash
4379 * to determine the slave interface -
4380 * (a) BOND_MODE_8023AD
4382 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4384 * The caller is expected to hold RTNL only and NO other lock!
4386 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4388 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4389 struct slave *slave;
4390 struct list_head *iter;
4396 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4397 bond->slave_cnt), GFP_KERNEL);
4398 all_slaves = kzalloc(struct_size(all_slaves, arr,
4399 bond->slave_cnt), GFP_KERNEL);
4400 if (!usable_slaves || !all_slaves) {
4404 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4405 struct ad_info ad_info;
4407 spin_lock_bh(&bond->mode_lock);
4408 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4409 spin_unlock_bh(&bond->mode_lock);
4410 pr_debug("bond_3ad_get_active_agg_info failed\n");
4411 /* No active aggragator means it's not safe to use
4412 * the previous array.
4414 bond_reset_slave_arr(bond);
4417 spin_unlock_bh(&bond->mode_lock);
4418 agg_id = ad_info.aggregator_id;
4420 bond_for_each_slave(bond, slave, iter) {
4421 if (skipslave == slave)
4424 all_slaves->arr[all_slaves->count++] = slave;
4425 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4426 struct aggregator *agg;
4428 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4429 if (!agg || agg->aggregator_identifier != agg_id)
4432 if (!bond_slave_can_tx(slave))
4435 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4436 usable_slaves->count);
4438 usable_slaves->arr[usable_slaves->count++] = slave;
4441 bond_set_slave_arr(bond, usable_slaves, all_slaves);
4444 if (ret != 0 && skipslave) {
4445 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4447 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4450 kfree_rcu(all_slaves, rcu);
4451 kfree_rcu(usable_slaves, rcu);
4456 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4457 struct sk_buff *skb,
4458 struct bond_up_slave *slaves)
4460 struct slave *slave;
4464 hash = bond_xmit_hash(bond, skb);
4465 count = slaves ? READ_ONCE(slaves->count) : 0;
4466 if (unlikely(!count))
4469 slave = slaves->arr[hash % count];
4473 /* Use this Xmit function for 3AD as well as XOR modes. The current
4474 * usable slave array is formed in the control path. The xmit function
4475 * just calculates hash and sends the packet out.
4477 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4478 struct net_device *dev)
4480 struct bonding *bond = netdev_priv(dev);
4481 struct bond_up_slave *slaves;
4482 struct slave *slave;
4484 slaves = rcu_dereference(bond->usable_slaves);
4485 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4487 return bond_dev_queue_xmit(bond, skb, slave->dev);
4489 return bond_tx_drop(dev, skb);
4492 /* in broadcast mode, we send everything to all usable interfaces. */
4493 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4494 struct net_device *bond_dev)
4496 struct bonding *bond = netdev_priv(bond_dev);
4497 struct slave *slave = NULL;
4498 struct list_head *iter;
4500 bond_for_each_slave_rcu(bond, slave, iter) {
4501 if (bond_is_last_slave(bond, slave))
4503 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4504 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4507 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4508 bond_dev->name, __func__);
4511 bond_dev_queue_xmit(bond, skb2, slave->dev);
4514 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4515 return bond_dev_queue_xmit(bond, skb, slave->dev);
4517 return bond_tx_drop(bond_dev, skb);
4520 /*------------------------- Device initialization ---------------------------*/
4522 /* Lookup the slave that corresponds to a qid */
4523 static inline int bond_slave_override(struct bonding *bond,
4524 struct sk_buff *skb)
4526 struct slave *slave = NULL;
4527 struct list_head *iter;
4529 if (!skb_rx_queue_recorded(skb))
4532 /* Find out if any slaves have the same mapping as this skb. */
4533 bond_for_each_slave_rcu(bond, slave, iter) {
4534 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4535 if (bond_slave_is_up(slave) &&
4536 slave->link == BOND_LINK_UP) {
4537 bond_dev_queue_xmit(bond, skb, slave->dev);
4540 /* If the slave isn't UP, use default transmit policy. */
4549 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4550 struct net_device *sb_dev)
4552 /* This helper function exists to help dev_pick_tx get the correct
4553 * destination queue. Using a helper function skips a call to
4554 * skb_tx_hash and will put the skbs in the queue we expect on their
4555 * way down to the bonding driver.
4557 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4559 /* Save the original txq to restore before passing to the driver */
4560 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4562 if (unlikely(txq >= dev->real_num_tx_queues)) {
4564 txq -= dev->real_num_tx_queues;
4565 } while (txq >= dev->real_num_tx_queues);
4570 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4571 struct sk_buff *skb,
4574 struct bonding *bond = netdev_priv(master_dev);
4575 struct bond_up_slave *slaves;
4576 struct slave *slave = NULL;
4578 switch (BOND_MODE(bond)) {
4579 case BOND_MODE_ROUNDROBIN:
4580 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4582 case BOND_MODE_ACTIVEBACKUP:
4583 slave = bond_xmit_activebackup_slave_get(bond, skb);
4585 case BOND_MODE_8023AD:
4588 slaves = rcu_dereference(bond->all_slaves);
4590 slaves = rcu_dereference(bond->usable_slaves);
4591 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4593 case BOND_MODE_BROADCAST:
4596 slave = bond_xmit_alb_slave_get(bond, skb);
4599 slave = bond_xmit_tlb_slave_get(bond, skb);
4602 /* Should never happen, mode already checked */
4603 WARN_ONCE(true, "Unknown bonding mode");
4612 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
4614 switch (sk->sk_family) {
4615 #if IS_ENABLED(CONFIG_IPV6)
4617 if (sk->sk_ipv6only ||
4618 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
4619 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
4620 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
4621 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
4626 default: /* AF_INET */
4627 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
4628 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
4629 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
4633 flow->ports.src = inet_sk(sk)->inet_sport;
4634 flow->ports.dst = inet_sk(sk)->inet_dport;
4638 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
4639 * @sk: socket to use for headers
4641 * This function will extract the necessary field from the socket and use
4642 * them to generate a hash based on the LAYER34 xmit_policy.
4643 * Assumes that sk is a TCP or UDP socket.
4645 static u32 bond_sk_hash_l34(struct sock *sk)
4647 struct flow_keys flow;
4650 bond_sk_to_flow(sk, &flow);
4653 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4655 return bond_ip_hash(hash, &flow);
4658 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
4661 struct bond_up_slave *slaves;
4662 struct slave *slave;
4666 slaves = rcu_dereference(bond->usable_slaves);
4667 count = slaves ? READ_ONCE(slaves->count) : 0;
4668 if (unlikely(!count))
4671 hash = bond_sk_hash_l34(sk);
4672 slave = slaves->arr[hash % count];
4677 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
4680 struct bonding *bond = netdev_priv(dev);
4681 struct net_device *lower = NULL;
4684 if (bond_sk_check(bond))
4685 lower = __bond_sk_get_lower_dev(bond, sk);
4691 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4692 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
4693 struct net_device *dev)
4695 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
4696 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
4697 return bond_tx_drop(dev, skb);
4701 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4703 struct bonding *bond = netdev_priv(dev);
4705 if (bond_should_override_tx_queue(bond) &&
4706 !bond_slave_override(bond, skb))
4707 return NETDEV_TX_OK;
4709 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4710 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
4711 return bond_tls_device_xmit(bond, skb, dev);
4714 switch (BOND_MODE(bond)) {
4715 case BOND_MODE_ROUNDROBIN:
4716 return bond_xmit_roundrobin(skb, dev);
4717 case BOND_MODE_ACTIVEBACKUP:
4718 return bond_xmit_activebackup(skb, dev);
4719 case BOND_MODE_8023AD:
4721 return bond_3ad_xor_xmit(skb, dev);
4722 case BOND_MODE_BROADCAST:
4723 return bond_xmit_broadcast(skb, dev);
4725 return bond_alb_xmit(skb, dev);
4727 return bond_tlb_xmit(skb, dev);
4729 /* Should never happen, mode already checked */
4730 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4732 return bond_tx_drop(dev, skb);
4736 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4738 struct bonding *bond = netdev_priv(dev);
4739 netdev_tx_t ret = NETDEV_TX_OK;
4741 /* If we risk deadlock from transmitting this in the
4742 * netpoll path, tell netpoll to queue the frame for later tx
4744 if (unlikely(is_netpoll_tx_blocked(dev)))
4745 return NETDEV_TX_BUSY;
4748 if (bond_has_slaves(bond))
4749 ret = __bond_start_xmit(skb, dev);
4751 ret = bond_tx_drop(dev, skb);
4757 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4759 if (speed == 0 || speed == SPEED_UNKNOWN)
4760 speed = slave->speed;
4762 speed = min(speed, slave->speed);
4767 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4768 struct ethtool_link_ksettings *cmd)
4770 struct bonding *bond = netdev_priv(bond_dev);
4771 struct list_head *iter;
4772 struct slave *slave;
4775 cmd->base.duplex = DUPLEX_UNKNOWN;
4776 cmd->base.port = PORT_OTHER;
4778 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4779 * do not need to check mode. Though link speed might not represent
4780 * the true receive or transmit bandwidth (not all modes are symmetric)
4781 * this is an accurate maximum.
4783 bond_for_each_slave(bond, slave, iter) {
4784 if (bond_slave_can_tx(slave)) {
4785 if (slave->speed != SPEED_UNKNOWN) {
4786 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4787 speed = bond_mode_bcast_speed(slave,
4790 speed += slave->speed;
4792 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4793 slave->duplex != DUPLEX_UNKNOWN)
4794 cmd->base.duplex = slave->duplex;
4797 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4802 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4803 struct ethtool_drvinfo *drvinfo)
4805 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4806 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4810 static const struct ethtool_ops bond_ethtool_ops = {
4811 .get_drvinfo = bond_ethtool_get_drvinfo,
4812 .get_link = ethtool_op_get_link,
4813 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4816 static const struct net_device_ops bond_netdev_ops = {
4817 .ndo_init = bond_init,
4818 .ndo_uninit = bond_uninit,
4819 .ndo_open = bond_open,
4820 .ndo_stop = bond_close,
4821 .ndo_start_xmit = bond_start_xmit,
4822 .ndo_select_queue = bond_select_queue,
4823 .ndo_get_stats64 = bond_get_stats,
4824 .ndo_do_ioctl = bond_do_ioctl,
4825 .ndo_change_rx_flags = bond_change_rx_flags,
4826 .ndo_set_rx_mode = bond_set_rx_mode,
4827 .ndo_change_mtu = bond_change_mtu,
4828 .ndo_set_mac_address = bond_set_mac_address,
4829 .ndo_neigh_setup = bond_neigh_setup,
4830 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4831 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4832 #ifdef CONFIG_NET_POLL_CONTROLLER
4833 .ndo_netpoll_setup = bond_netpoll_setup,
4834 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4835 .ndo_poll_controller = bond_poll_controller,
4837 .ndo_add_slave = bond_enslave,
4838 .ndo_del_slave = bond_release,
4839 .ndo_fix_features = bond_fix_features,
4840 .ndo_features_check = passthru_features_check,
4841 .ndo_get_xmit_slave = bond_xmit_get_slave,
4842 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
4845 static const struct device_type bond_type = {
4849 static void bond_destructor(struct net_device *bond_dev)
4851 struct bonding *bond = netdev_priv(bond_dev);
4853 destroy_workqueue(bond->wq);
4856 void bond_setup(struct net_device *bond_dev)
4858 struct bonding *bond = netdev_priv(bond_dev);
4860 spin_lock_init(&bond->mode_lock);
4861 bond->params = bonding_defaults;
4863 /* Initialize pointers */
4864 bond->dev = bond_dev;
4866 /* Initialize the device entry points */
4867 ether_setup(bond_dev);
4868 bond_dev->max_mtu = ETH_MAX_MTU;
4869 bond_dev->netdev_ops = &bond_netdev_ops;
4870 bond_dev->ethtool_ops = &bond_ethtool_ops;
4872 bond_dev->needs_free_netdev = true;
4873 bond_dev->priv_destructor = bond_destructor;
4875 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4877 /* Initialize the device options */
4878 bond_dev->flags |= IFF_MASTER;
4879 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4880 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4882 #ifdef CONFIG_XFRM_OFFLOAD
4883 /* set up xfrm device ops (only supported in active-backup right now) */
4884 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
4886 #endif /* CONFIG_XFRM_OFFLOAD */
4888 /* don't acquire bond device's netif_tx_lock when transmitting */
4889 bond_dev->features |= NETIF_F_LLTX;
4891 /* By default, we declare the bond to be fully
4892 * VLAN hardware accelerated capable. Special
4893 * care is taken in the various xmit functions
4894 * when there are slaves that are not hw accel
4898 /* Don't allow bond devices to change network namespaces. */
4899 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4901 bond_dev->hw_features = BOND_VLAN_FEATURES |
4902 NETIF_F_HW_VLAN_CTAG_RX |
4903 NETIF_F_HW_VLAN_CTAG_FILTER;
4905 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4906 bond_dev->features |= bond_dev->hw_features;
4907 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4908 #ifdef CONFIG_XFRM_OFFLOAD
4909 bond_dev->hw_features |= BOND_XFRM_FEATURES;
4910 /* Only enable XFRM features if this is an active-backup config */
4911 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4912 bond_dev->features |= BOND_XFRM_FEATURES;
4913 #endif /* CONFIG_XFRM_OFFLOAD */
4914 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4915 if (bond_sk_check(bond))
4916 bond_dev->features |= BOND_TLS_FEATURES;
4920 /* Destroy a bonding device.
4921 * Must be under rtnl_lock when this function is called.
4923 static void bond_uninit(struct net_device *bond_dev)
4925 struct bonding *bond = netdev_priv(bond_dev);
4926 struct bond_up_slave *usable, *all;
4927 struct list_head *iter;
4928 struct slave *slave;
4930 bond_netpoll_cleanup(bond_dev);
4932 /* Release the bonded slaves */
4933 bond_for_each_slave(bond, slave, iter)
4934 __bond_release_one(bond_dev, slave->dev, true, true);
4935 netdev_info(bond_dev, "Released all slaves\n");
4937 usable = rtnl_dereference(bond->usable_slaves);
4939 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4940 kfree_rcu(usable, rcu);
4943 all = rtnl_dereference(bond->all_slaves);
4945 RCU_INIT_POINTER(bond->all_slaves, NULL);
4946 kfree_rcu(all, rcu);
4949 list_del(&bond->bond_list);
4951 bond_debug_unregister(bond);
4954 /*------------------------- Module initialization ---------------------------*/
4956 static int bond_check_params(struct bond_params *params)
4958 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4959 struct bond_opt_value newval;
4960 const struct bond_opt_value *valptr;
4961 int arp_all_targets_value = 0;
4962 u16 ad_actor_sys_prio = 0;
4963 u16 ad_user_port_key = 0;
4964 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4966 int bond_mode = BOND_MODE_ROUNDROBIN;
4967 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4971 /* Convert string parameters. */
4973 bond_opt_initstr(&newval, mode);
4974 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4976 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4979 bond_mode = valptr->value;
4982 if (xmit_hash_policy) {
4983 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4984 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4985 bond_mode == BOND_MODE_BROADCAST) {
4986 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4987 bond_mode_name(bond_mode));
4989 bond_opt_initstr(&newval, xmit_hash_policy);
4990 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4993 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4997 xmit_hashtype = valptr->value;
5002 if (bond_mode != BOND_MODE_8023AD) {
5003 pr_info("lacp_rate param is irrelevant in mode %s\n",
5004 bond_mode_name(bond_mode));
5006 bond_opt_initstr(&newval, lacp_rate);
5007 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5010 pr_err("Error: Invalid lacp rate \"%s\"\n",
5014 lacp_fast = valptr->value;
5019 bond_opt_initstr(&newval, ad_select);
5020 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5023 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5026 params->ad_select = valptr->value;
5027 if (bond_mode != BOND_MODE_8023AD)
5028 pr_warn("ad_select param only affects 802.3ad mode\n");
5030 params->ad_select = BOND_AD_STABLE;
5033 if (max_bonds < 0) {
5034 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5035 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5036 max_bonds = BOND_DEFAULT_MAX_BONDS;
5040 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5046 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5051 if (downdelay < 0) {
5052 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5053 downdelay, INT_MAX);
5057 if ((use_carrier != 0) && (use_carrier != 1)) {
5058 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5063 if (num_peer_notif < 0 || num_peer_notif > 255) {
5064 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5069 /* reset values for 802.3ad/TLB/ALB */
5070 if (!bond_mode_uses_arp(bond_mode)) {
5072 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
5073 pr_warn("Forcing miimon to 100msec\n");
5074 miimon = BOND_DEFAULT_MIIMON;
5078 if (tx_queues < 1 || tx_queues > 255) {
5079 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5080 tx_queues, BOND_DEFAULT_TX_QUEUES);
5081 tx_queues = BOND_DEFAULT_TX_QUEUES;
5084 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5085 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5087 all_slaves_active = 0;
5090 if (resend_igmp < 0 || resend_igmp > 255) {
5091 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5092 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5093 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5096 bond_opt_initval(&newval, packets_per_slave);
5097 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5098 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5099 packets_per_slave, USHRT_MAX);
5100 packets_per_slave = 1;
5103 if (bond_mode == BOND_MODE_ALB) {
5104 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
5109 if (updelay || downdelay) {
5110 /* just warn the user the up/down delay will have
5111 * no effect since miimon is zero...
5113 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
5114 updelay, downdelay);
5117 /* don't allow arp monitoring */
5119 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5120 miimon, arp_interval);
5124 if ((updelay % miimon) != 0) {
5125 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5126 updelay, miimon, (updelay / miimon) * miimon);
5131 if ((downdelay % miimon) != 0) {
5132 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5134 (downdelay / miimon) * miimon);
5137 downdelay /= miimon;
5140 if (arp_interval < 0) {
5141 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5142 arp_interval, INT_MAX);
5146 for (arp_ip_count = 0, i = 0;
5147 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5150 /* not a complete check, but good enough to catch mistakes */
5151 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5152 !bond_is_ip_target_ok(ip)) {
5153 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5157 if (bond_get_targets_ip(arp_target, ip) == -1)
5158 arp_target[arp_ip_count++] = ip;
5160 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5165 if (arp_interval && !arp_ip_count) {
5166 /* don't allow arping if no arp_ip_target given... */
5167 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5173 if (!arp_interval) {
5174 pr_err("arp_validate requires arp_interval\n");
5178 bond_opt_initstr(&newval, arp_validate);
5179 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5182 pr_err("Error: invalid arp_validate \"%s\"\n",
5186 arp_validate_value = valptr->value;
5188 arp_validate_value = 0;
5191 if (arp_all_targets) {
5192 bond_opt_initstr(&newval, arp_all_targets);
5193 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5196 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5198 arp_all_targets_value = 0;
5200 arp_all_targets_value = valptr->value;
5205 pr_info("MII link monitoring set to %d ms\n", miimon);
5206 } else if (arp_interval) {
5207 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5208 arp_validate_value);
5209 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5210 arp_interval, valptr->string, arp_ip_count);
5212 for (i = 0; i < arp_ip_count; i++)
5213 pr_cont(" %s", arp_ip_target[i]);
5217 } else if (max_bonds) {
5218 /* miimon and arp_interval not set, we need one so things
5219 * work as expected, see bonding.txt for details
5221 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
5224 if (primary && !bond_mode_uses_primary(bond_mode)) {
5225 /* currently, using a primary only makes sense
5226 * in active backup, TLB or ALB modes
5228 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5229 primary, bond_mode_name(bond_mode));
5233 if (primary && primary_reselect) {
5234 bond_opt_initstr(&newval, primary_reselect);
5235 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5238 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5242 primary_reselect_value = valptr->value;
5244 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5247 if (fail_over_mac) {
5248 bond_opt_initstr(&newval, fail_over_mac);
5249 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5252 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5256 fail_over_mac_value = valptr->value;
5257 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5258 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5260 fail_over_mac_value = BOND_FOM_NONE;
5263 bond_opt_initstr(&newval, "default");
5264 valptr = bond_opt_parse(
5265 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5268 pr_err("Error: No ad_actor_sys_prio default value");
5271 ad_actor_sys_prio = valptr->value;
5273 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5276 pr_err("Error: No ad_user_port_key default value");
5279 ad_user_port_key = valptr->value;
5281 bond_opt_initstr(&newval, "default");
5282 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5284 pr_err("Error: No tlb_dynamic_lb default value");
5287 tlb_dynamic_lb = valptr->value;
5289 if (lp_interval == 0) {
5290 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5291 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5292 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5295 /* fill params struct with the proper values */
5296 params->mode = bond_mode;
5297 params->xmit_policy = xmit_hashtype;
5298 params->miimon = miimon;
5299 params->num_peer_notif = num_peer_notif;
5300 params->arp_interval = arp_interval;
5301 params->arp_validate = arp_validate_value;
5302 params->arp_all_targets = arp_all_targets_value;
5303 params->updelay = updelay;
5304 params->downdelay = downdelay;
5305 params->peer_notif_delay = 0;
5306 params->use_carrier = use_carrier;
5307 params->lacp_fast = lacp_fast;
5308 params->primary[0] = 0;
5309 params->primary_reselect = primary_reselect_value;
5310 params->fail_over_mac = fail_over_mac_value;
5311 params->tx_queues = tx_queues;
5312 params->all_slaves_active = all_slaves_active;
5313 params->resend_igmp = resend_igmp;
5314 params->min_links = min_links;
5315 params->lp_interval = lp_interval;
5316 params->packets_per_slave = packets_per_slave;
5317 params->tlb_dynamic_lb = tlb_dynamic_lb;
5318 params->ad_actor_sys_prio = ad_actor_sys_prio;
5319 eth_zero_addr(params->ad_actor_system);
5320 params->ad_user_port_key = ad_user_port_key;
5321 if (packets_per_slave > 0) {
5322 params->reciprocal_packets_per_slave =
5323 reciprocal_value(packets_per_slave);
5325 /* reciprocal_packets_per_slave is unused if
5326 * packets_per_slave is 0 or 1, just initialize it
5328 params->reciprocal_packets_per_slave =
5329 (struct reciprocal_value) { 0 };
5333 strncpy(params->primary, primary, IFNAMSIZ);
5334 params->primary[IFNAMSIZ - 1] = 0;
5337 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5342 /* Called from registration process */
5343 static int bond_init(struct net_device *bond_dev)
5345 struct bonding *bond = netdev_priv(bond_dev);
5346 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5348 netdev_dbg(bond_dev, "Begin bond_init\n");
5350 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5354 spin_lock_init(&bond->stats_lock);
5355 netdev_lockdep_set_classes(bond_dev);
5357 list_add_tail(&bond->bond_list, &bn->dev_list);
5359 bond_prepare_sysfs_group(bond);
5361 bond_debug_register(bond);
5363 /* Ensure valid dev_addr */
5364 if (is_zero_ether_addr(bond_dev->dev_addr) &&
5365 bond_dev->addr_assign_type == NET_ADDR_PERM)
5366 eth_hw_addr_random(bond_dev);
5371 unsigned int bond_get_num_tx_queues(void)
5376 /* Create a new bond based on the specified name and bonding parameters.
5377 * If name is NULL, obtain a suitable "bond%d" name for us.
5378 * Caller must NOT hold rtnl_lock; we need to release it here before we
5379 * set up our sysfs entries.
5381 int bond_create(struct net *net, const char *name)
5383 struct net_device *bond_dev;
5384 struct bonding *bond;
5385 struct alb_bond_info *bond_info;
5390 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5391 name ? name : "bond%d", NET_NAME_UNKNOWN,
5392 bond_setup, tx_queues);
5394 pr_err("%s: eek! can't alloc netdev!\n", name);
5400 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5401 * It is set to 0 by default which is wrong.
5403 bond = netdev_priv(bond_dev);
5404 bond_info = &(BOND_ALB_INFO(bond));
5405 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5407 dev_net_set(bond_dev, net);
5408 bond_dev->rtnl_link_ops = &bond_link_ops;
5410 res = register_netdevice(bond_dev);
5412 free_netdev(bond_dev);
5418 netif_carrier_off(bond_dev);
5420 bond_work_init_all(bond);
5426 static int __net_init bond_net_init(struct net *net)
5428 struct bond_net *bn = net_generic(net, bond_net_id);
5431 INIT_LIST_HEAD(&bn->dev_list);
5433 bond_create_proc_dir(bn);
5434 bond_create_sysfs(bn);
5439 static void __net_exit bond_net_exit(struct net *net)
5441 struct bond_net *bn = net_generic(net, bond_net_id);
5442 struct bonding *bond, *tmp_bond;
5445 bond_destroy_sysfs(bn);
5447 /* Kill off any bonds created after unregistering bond rtnl ops */
5449 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5450 unregister_netdevice_queue(bond->dev, &list);
5451 unregister_netdevice_many(&list);
5454 bond_destroy_proc_dir(bn);
5457 static struct pernet_operations bond_net_ops = {
5458 .init = bond_net_init,
5459 .exit = bond_net_exit,
5461 .size = sizeof(struct bond_net),
5464 static int __init bonding_init(void)
5469 res = bond_check_params(&bonding_defaults);
5473 res = register_pernet_subsys(&bond_net_ops);
5477 res = bond_netlink_init();
5481 bond_create_debugfs();
5483 for (i = 0; i < max_bonds; i++) {
5484 res = bond_create(&init_net, NULL);
5489 skb_flow_dissector_init(&flow_keys_bonding,
5490 flow_keys_bonding_keys,
5491 ARRAY_SIZE(flow_keys_bonding_keys));
5493 register_netdevice_notifier(&bond_netdev_notifier);
5497 bond_destroy_debugfs();
5498 bond_netlink_fini();
5500 unregister_pernet_subsys(&bond_net_ops);
5505 static void __exit bonding_exit(void)
5507 unregister_netdevice_notifier(&bond_netdev_notifier);
5509 bond_destroy_debugfs();
5511 bond_netlink_fini();
5512 unregister_pernet_subsys(&bond_net_ops);
5514 #ifdef CONFIG_NET_POLL_CONTROLLER
5515 /* Make sure we don't have an imbalance on our netpoll blocking */
5516 WARN_ON(atomic_read(&netpoll_block_tx));
5520 module_init(bonding_init);
5521 module_exit(bonding_exit);
5522 MODULE_LICENSE("GPL");
5523 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5524 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");